AI Tools for Creatives – January to March 2026

The first quarter of 2026 was defined by acceleration on every front: frontier models iterated at velocity (267 AI models released in Q1 alone), agentic systems graduated from demos to default workflows, and the race between closed and open-source closed noticeably. For creatives, the signal-to-noise improved — image quality jumped with Recraft V4, Nano Banana 2, and a near-release Midjourney V8; video got natively cinematic with Kling 3.0 and Seedance 2.0; and audio finally felt expressive with ElevenLabs Eleven v3. The platform consolidation story also intensified: Adobe merged Acrobat and Express, Anthropic acquired Vercept, and ElevenLabs repositioned itself as a full-stack Audio OS.

General AI Tools

ChatGPT / OpenAI — (Major Update) Three significant model releases in the period. GPT-5.3-Codex (Feb 5) is OpenAI's most capable agentic coding model, 25% faster than its predecessor and rated "High" for cybersecurity risk on their own preparedness framework. GPT-5.3-Codex-Spark (Feb 12) is a real-time coding variant delivering >1,000 tokens/second on Cerebras hardware, designed for interactive use alongside the main Codex model. GPT-5.4 (Mar 5) is the headline release: a 272K–1.05M token context window, a full Computer Use API (surpassing human baseline on OSWorld-Verified at 75%), configurable reasoning effort across five levels, and 33% fewer factual errors than GPT-5.2. Alongside the model cadence, OpenAI launched a new $8/month Go tier, the Prism scientific workspace, and began a phased US-only rollout of ads in ChatGPT.

Claude (Anthropic) — (Major Update) Claude Opus 3 was retired in January — the first Anthropic model to go through a formal, publicly documented retirement process. Opus 4.6 (Feb 5) followed with a 1M-token context window in beta, Agent Teams (multiple Claude instances coordinating in Claude Code), and direct editing inside PowerPoint and Excel. Sonnet 4.6 (Feb 17) became the new default for free and Pro users, hitting 72.5% on OSWorld computer-use benchmarks. Anthropic also acquired Vercept (Feb 25), a Seattle computer-use startup whose VyUI technology strengthens Claude's ability to operate desktop interfaces. The company reached a $380B valuation on $30B Series G funding and grew enterprise AI spend share from ~4% to 40% in a year — partly because it publicly declined to grant autonomous weapons access to the US government.

Perplexity — (Major Update) The period transformed Perplexity from a research tool into something closer to a multi-model operating system. Model Council (Feb 6) lets users cross-check answers across GPT-5.4, Claude Opus 4.6, and Gemini 3.1 Pro simultaneously, synthesising agreements and contradictions. Perplexity Computer (Feb 27) unifies 19 AI models in one conversation for planning, delegation, and multi-agent project execution. Personal Computer (Mar 12) extends this to a user's local Mac mini, giving the cloud AI persistent access to files, apps, and sessions. Memory recall improved from 77% to 95%. The company eliminated advertising entirely in favour of a subscription-only model.

Gemini (Google) — (Major Update) Gemini 3.1 Pro launched February 19, dominating 13 of 16 major benchmarks at release and achieving 77.1% on ARC-AGI-2 — more than double previous performance. It replaced Gemini 3 Pro Preview entirely by March 9. In January, Google rolled out Personal Intelligence (connecting Gemini to Gmail, Photos, and YouTube history), Agentic Vision in Gemini 3 Flash (active image exploration rather than passive viewing), and full Gemini integration into Chrome for multi-step task execution. March 10 brought Gemini capabilities directly into Docs, Sheets, Slides, and Drive — generating formatted first drafts from existing content across the Workspace ecosystem without switching apps.

Microsoft Copilot — (Major Update) Agent Mode rolled out across Word, Excel, and PowerPoint throughout the quarter — the AI now edits documents while surfacing its reasoning, rather than just suggesting text. The Project Manager Agent (March, public preview) plans work, integrates with Planner and Teams, and pulls decisions from meeting transcripts to generate task lists. GPT-5.3 Instant (Mar 3) and GPT-5.4 Thinking (Mar 6) both landed in M365 Copilot Chat, with the latter enabling deep reasoning for complex agentic workflows. Multi-agent coordination — where Copilot agents call other agents as tools — began rolling out in March.

ChatGPT Atlas — (Updated) OpenAI's macOS browser continued a steady cadence of weekly builds. Highlights include Tab Groups with titles and emoji (Jan 21), Saved Prompts bookmarkable via "@" reference (Feb 4), Auto Organize tabs using ChatGPT memory (Feb 18), reduced Agent Mode "laziness" on repetitive tasks like bulk email processing (Feb 24), and multi-account login support (Mar 10). Still macOS-only; no Windows or mobile release announced.

ChatGPT Pulse — (Still Relevant) No new Pulse-specific features launched in the period. The daily personalised research briefing remains restricted to Pro subscribers ($200/month); no Plus rollout date confirmed despite OpenAI stating it's the goal "when it can be made more efficient."

DeepSeek V4 — (New) Released around March 3, DeepSeek V4 is a 1-trillion-parameter sparse MoE model with 32B active parameters per token. Technical highlights include a tiered KV cache cutting memory use by 40%, Sparse FP8 decoding (1.8x speedup), and the Engram architecture for efficient retrieval in 1M+ token contexts. Open-weight and natively multimodal, it remains competitive with proprietary frontier models at a fraction of the API cost.

Grok 4.20 (xAI) — (New) Launched February 17 alongside Claude Sonnet 4.6 and Gemini 3.1 Pro in an unusually busy single day for frontier releases. Grok 4.20 introduces a four-agent architecture for complex task decomposition. Limited public technical detail from xAI beyond the multi-agent capability.

Creative AI Tools

Adobe Firefly — (Major Update) Photoshop's Generative Fill, Generative Expand, and Remove Tool all upgraded on January 27 to deliver 2K resolution output with sharper detail, fewer artefacts, better prompt matching, and more natural lighting — powered by the new Firefly Fill & Expand and Remove Tool 3 models. On March 11, Quick Cut arrived in beta within the Firefly Video Editor: it assembles uploaded footage into a structured first-draft edit from a text prompt, detecting scene changes, identifying key dialogue cues, and arranging clips into a narrative sequence. Third-party model access expanded further with FLUX.2 (Black Forest Labs), Nano Banana Pro, and GPT Image now all available as partner models inside Firefly.

Adobe Acrobat Studio — (Major Update) On January 21, Adobe merged Acrobat and Adobe Express into a single product called Acrobat Studio. New AI features at launch include: natural-language PDF editing ("delete page 3, add a password"), AI-generated presentations from documents and URLs, a Generate Podcast feature that turns collections of PDFs and links into audio summaries, and AI-powered highlights that learn from reading behaviour. AI use across Acrobat was up 4× year-on-year at launch. The $1.9B Semrush acquisition, expected to close H1 2026, adds competitive intelligence to the AI marketing suite.

Midjourney — (Major Update) Niji 7 launched January 9 — the first Niji update since January 2024 — bringing significantly improved coherency, clean line work, and vibrant colour output for anime-style generation. V8 remained in final testing for the entire period: a full codebase rewrite with new APIs, optional 2K resolution mode, better text rendering, and improved body coherence. The final ranking party completed February 20, and by the March 4–8 Office Hours V8 was described as "launchable as soon as next week" but held back for timing. Personalisation also received a major UI overhaul in February. V8 had not launched as of March 17.

Recraft — (Major Update) Recraft V4 launched February 17 as a ground-up rebuild focused on "design taste" — compositional judgment, lighting, material realism, and colour relationships co-developed with professional designers. V4 Pro renders at 2048×2048 for print-ready output. The vector generation capability remains unique: Recraft is the only AI tool producing true editable SVG files with clean geometry and structured layers, no tracing required. Agentic Mode lets you refine designs via natural-language conversation on an infinite canvas, with access to Nano Banana, GPT Image, and Flux as external models. Available on free and paid plans.

Google Nano Banana 2 (Gemini 3.1 Flash Image) — (Major Update) Nano Banana 2 launched February 26, combining Nano Banana Pro's visual quality with Gemini Flash's generation speed. New capabilities include advanced world knowledge drawn from real-time web search (accurately rendering specific subjects), data visualisation from notes or structured data, precision text rendering with in-image translation, and subject consistency across edits. Rolling out across Gemini app, Google Search, and Google Ads. All outputs carry SynthID + C2PA Content Credentials.

DALL·E / GPT Image — (Major Update) DALL-E 3 was effectively replaced in ChatGPT by GPT Image 1.5 (launched December 16, 2025), which delivers up to 4× faster generation, precise edits that preserve lighting and facial likeness, improved dense text rendering, and a dedicated Images sidebar in ChatGPT with preset filters and one-time likeness upload. DALL-E 2 and DALL-E 3 API snapshots are both scheduled for deprecation on May 12, 2026; OpenAI recommends migrating to gpt-image-1 or gpt-image-1-mini.

Ideogram — (Still Relevant) Ideogram 3.0 (launched March 2025) continued as the go-to tool for text-in-images, maintaining ~90% text rendering accuracy against Midjourney's ~30% for short phrases. In community benchmarks from February 2026, Ideogram still rated 10/10 for text vs. Recraft V3 (8/10) and Nano Banana (7/10). Multi-modal AI (text + image + audio prompts), expanded style libraries (100+), and improved processing speed are noted for 2026 broadly, though no single major version launch is confirmed for Jan–Mar.

Leonardo AI — (Updated) A Dynamic Lighting update to Lucid models arrived in January 2026, improving shadow realism and reducing the "plastic" look in portraits. The core platform otherwise continued with Phoenix (high-fidelity generation with strong prompt adherence), Alchemy v4's Hyper-Realism and Abstract Concept modes, Motion v3 for 10-second HD video clips, and Real-Time Canvas for composite editing. Canva Business ($20/month) now includes Leonardo access.

Flora — (Major Update) Node-based creative platform Flora raised $42M in a Series A led by Redpoint Ventures on January 27 (total funding $52M), with clients including Lionsgate, Levi's, and Pentagram. The Organic Composition Engine — placing objects naturally in scenes with depth and weight understanding — launched in 2026. The Multi-Model Hub provides access to Flux 2 Pro, Seedream 4.5, Veo 3, Nano Banana, and GPT-5.1 in a single node-based canvas, without separate subscriptions. Style DNA lets teams train custom LoRAs from 15–20 reference images to generate on-brand variations at scale.

Higgsfield — (Major Update) Substantial feature expansion throughout the quarter. Relight launched January 6 with directional 3D light control, softbox/hard-light toggle, and depth-mapping for realistic relighting of photos. Cinema Studio 2.0 (Feb 26) added the "What's Next" narrative suggestion feature and more granular camera controls for commercial-grade video storytelling. March brought Higgsfield Audio (voiceover and voice-change functions), Vibe Motion in beta (SVG/PNG assets animated into CSS-based motion graphics), and the Similarity Scoring Tool (Mar 13) — achieving 86.6% accuracy flagging AI content for likeness of celebrities, brand logos, and cinematic signatures.

IC Light V2 — (Updated) IC Light V2 (by ControlNet creator Lvmin Zhang, presented at ICLR 2025) became available as a REST inference API on WaveSpeedAI on January 29, and is also accessible via fal.ai. V2 improvements over V1 include advanced 16-channel VAE technology for more powerful lighting control, text-driven relighting via natural-language descriptions, and five directional light options. Non-destructive transformations preserve detail throughout.

Magnific — (Still Relevant) Consistently ranked #1 for AI art upscaling in third-party benchmarks through early 2026. No major new feature launch identified in the period; the platform's "Creativity" slider and prompt-guided enhancement at up to 16× remain the differentiator for Midjourney and Stable Diffusion post-processing.

Krea — (Still Relevant) Real-time canvas and image/video enhancement suite remains operational, though community reports from early February flagged quality degradation post-update — yellow/warm tinting, increased graininess, and blurring. Krea dropped to #6 in Curious Refuge's 2026 upscaler rankings. No major feature release in the period.

FLUX.2 [klein] (Black Forest Labs) — (New) Open-source image generation model released January 16 with sub-second generation on modern hardware. The 4B parameter Apache 2.0 variant runs on ~13GB VRAM (consumer RTX 4070), supports multi-reference editing from up to 10 reference images, and uses 4-step inference. Available on Hugging Face, Cloudflare Workers AI, fal.ai, and RunDiffusion. Now integrated into Adobe Firefly and Flora.

Grok Imagine 1.0 (xAI) — (New) xAI's video generation platform launched January 28 (API) and February 3 (product), generating 10-second 720p video at $0.05/second with improved audio, sharper motion, and an "Extend from Frame" feature (March 2) for chaining clips. 1.245 billion videos were generated in January 2026 alone. Under regulatory scrutiny from both the UK ICO and Ireland's DPC over deepfake and real-person image concerns.

Autodesk Wonder 3D — (New) Launched March 4 within Autodesk Flow Studio, Wonder 3D generates fully editable 3D assets from text prompts or images (sketches, concept art). Outputs include geometry and textures, exported as .OBJ. Available to all Autodesk Flow Studio subscription tiers.

Presti — (Still Relevant) Paris-based furniture visualisation AI now serves 400+ furniture brands and is rapidly expanding into the US market. Real-time collaboration tools were added in the period. No major model update confirmed; entry plan at $330/month remains significantly higher than comparable tools.

Botika — (Still Relevant) AI fashion model generator for e-commerce continues to grow; no major product update announced in the period, though Botika was cited in March industry guides as part of the recommended AI visual stack for fashion brands.

Stable Virtual Camera (Stability AI) — (Still Relevant) SEVA v1.1 (June 2025) remains the current version with no new release in the period. The tool generates 3D-consistent novel views and dynamic camera paths (360°, spiral, dolly zoom, etc.) from 1–32 input images, free for research under a Non-Commercial License.

Story Diffusion — (Still Relevant) No confirmed product update in the period. The platform continues to offer character-consistent story sequence generation from photo references; no significant changelog or feature announcement found.

Khroma — (Still Relevant) AI colour palette generator that learns preferences from an initial 50-colour selection. No product updates found in the period; continues to be referenced in design tool roundups.

Fontjoy — (Still Relevant) Neural-network font pairing tool. No updates in the period; remains a useful single-purpose utility for rapidly generating harmonious typeface combinations.

AutoDraw (Google) — (Still Relevant) Sketch-prediction drawing tool from Google. No updates in the period; predates the generative AI wave and has not been publicly updated in several years.

Video AI Tools

Veo 3.1 (Google) — (Updated) A January 13 update brought three improvements to Veo 3.1: native vertical (9:16) video output for YouTube Shorts and social platforms via the Ingredients to Video feature (a first), improved expressiveness in dialogue and character consistency, and state-of-the-art upscaling to 1080p and 4K. Available in the Gemini app, YouTube Create, Google Vids, Flow, and Vertex AI. As of March 2026, Veo 3.1 commands an estimated 96.4% market share among production video users. Veo 4 remains unconfirmed; community expectation is a Google I/O announcement in May.

Runway — (Major Update) Gen-4.5 (launched December 2025) was the dominant model throughout Q1, holding the #1 position on the Artificial Analysis Text to Video benchmark with 1,247 Elo points — above Sora 2 and Veo 3.1. It offers precise camera direction from prompts, complex multi-shot support, and superior physical accuracy (liquids, hair, materials). Aleph, Runway's in-context video editing model, continued its rolling Enterprise/Creative Partners deployment in Q1 — it enables text-based object addition, removal, transformation, style transfer, and lighting adjustment on existing footage. A "significant update" was signalled for April 2026.

Kling 3.0 (Kuaishou) — (Major Update) Kling 3.0 launched February 5, three days before Seedance 2.0 in a clearly competitive timing decision. It introduces a Multi-modal Visual Language (MVL) framework processing text, images, audio, and video in one unified system. Key additions: native multi-language lip-synced audio across English, Chinese, Japanese, Korean, and Spanish; a multi-shot storyboard tool for directing duration, angle, pacing, and camera movement per shot; 4K image output; Motion Brush for manually painting motion paths on frames; and strong character consistency across shots. 30fps standard, 60fps at Ultra tier. The storyboard-to-sequence capability represents a fundamental shift from single-shot generation to director-style scene construction.

Seedance 2.0 (ByteDance) — (Major Update) ByteDance's Seedance 2.0 (February 8) arrives as a multimodal leap: text, image, audio, and video inputs processed in a single unified architecture that generates dialogue, ambient sounds, and music natively alongside the video. It accepts up to 12 reference files — the broadest input breadth in the category — and enables camera/motion replication from a reference video. Strong character, clothing, text, and scene consistency across frames addresses the character drift problem that has plagued generative video. API cost on Atlas Cloud is $0.022/second, significantly undercutting Kling 3.0 ($0.126/sec) and Sora 2 ($0.15/sec).

Sora (OpenAI) — (Updated) Sora was fully shut down due to not beeing profitable.

Luma AI / Dream Machine (Ray3.14) — (Major Update) Ray3.14 launched January 26 as a major upgrade targeting professional workflows: native 1080p across core Dream Machine workflows (no post-upscaling), 4× faster generation speed versus Ray3, 3× cheaper per second, and best-ever quality for animation and video-to-video. Modify Video now supports up to 18 seconds with improved consistency. Luma also announced a $1M Cannes Lions creative challenge in February. References (Character) not yet supported in Ray3.14.

Decart AI / Mirage (Lucy 2.0) — (Major Update) Lucy 2.0 launched January 26 as a real-time world transformation model generating 1080p at 30fps with under 40ms latency — running indefinitely without buffering or time constraints. Operating cost dropped from several hundred dollars/hour to ~$3/hour, with an API at ~$0.05/second. Characters, clothing, lighting, and object interactions remain consistent in live scenarios; the technology has been tested with Twitch streamers for real-time character and environment swaps. Forbes described it as "GPT-3 moment for world models." Lucy-14B (also active in Q1) produces 5-second clips in ~6.5 seconds — 7× faster than the baseline.

HeyGen — (Major Update) Two substantial releases in consecutive months. January: avatar creation rebuilt to a 15-second webcam workflow; Video Agent 2.0 shows a complete creative blueprint before rendering; a new Business plan bundles 5× capacity with 60-minute video/translation support and access to Avatar IV, Sora 2, and Veo 3. February: ChatGPT integration enables "describe video in chat → finished HeyGen video" via the Video Agent; a Video Agent API landed with Claude Code integration via HeyGen Skills; Avatar Memory saves high-quality motion clips for reuse with 50% faster generation. Avatar IV renders at 4K natively for Business/Enterprise.

Synthesia — (Major Update) Synthesia raised $200M at a $4B valuation (Google Ventures–led, January 26) and pivoted its strategic focus toward AI-powered employee training and conversational video agents. Synthesia 3.0 features now active in Q1 include Express-2 Avatars with natural body movement, Action-capable Avatars for B-roll prompts ("walk to the whiteboard"), voice cloning in 160+ languages, Interactivity 2.0 (quizzes, branching scenarios), AI Dubbing in 30+ languages, and an AI Playground with Sora 2, Veo 3.1, Nano Banana Pro, and FLUX 2 inside the editor.

Descript — (Updated) Underlord, Descript's AI co-editor, received significant upgrades in Q1. New features include smart filling layers (auto-populating text from speaker audio), in-app image generation from prompts, AI video clip generation from reference images, multi-step AI Templates for automating entire editing pipelines, and the integration of Kling Avatar v2 for greatly improved AI avatar generation. The core text-based editing workflow (edit transcript → edit video) now has a fuller generative AI layer alongside it.

Filmora AI / Wondershare — (Major Update) Filmora V15 with AI Mate launched for Q1 2026 — an intelligent editing assistant with four modes: AIGC (scripts and concept ideation), Guide (real-time workflow help), Action (multi-step operation execution in one click), and Auto (selects the right mode from intent). V15 integrates Nano Banana Pro, Sora 2, and Veo 3.1 directly into the editing pipeline for scene-to-scene transitions and AI content generation. AI timeline extension generates extra frames matching original motion and lighting.

Pika (Pika Labs) — (Updated) Pika 2.5 continued as the active engine throughout Q1. New experience-layer features include Pikaformance (still image to lifelike performance using a voice), Pika Selves (personalised digital avatar), and ongoing Pikatwists (video-to-video transformation). Sound effect generation and upgraded lipsync for complex expressions are live. Pika 2.5 generates at 1080p with ~42 second average render time; strong community reputation for physics-based effects and social-native content.

Moonvalley Marey — (Still Relevant) No new model release in the period; Marey's positioning as the "world's first commercially safe video model" (trained entirely on licensed footage) continues to distinguish it in enterprise and Hollywood contexts. Studios are actively using Marey for B-roll and background footage. Available on fal.ai and via ComfyUI.

Viggle — (Still Relevant) AI motion-capture platform for animating character images using motion reference videos, primarily for viral meme content. No major feature release in Q1 2026; the platform published a blog post noting "motion is the default" for memes in 2026.

Revid — (Still Relevant) Text/script-to-short-form video platform with 2,693+ TikTok and 2,099+ YouTube Shorts templates, auto captions in 100+ languages, multi-platform aspect ratio export in a single project, and hook generator for first-3-second attention. No major new feature release announced in Q1 2026; ongoing refinement of existing AI suite.

Riverside — (Still Relevant) Video podcast recording and AI-powered post-production suite. Magic Clips, text-based editing, Auto Translation with Lip Sync, and multi-aspect-ratio export actively used throughout Q1. No major new feature announcements; the platform was heavily promoted by creators as a "one recording → week of content" solution.

Arcads — (Updated) Veo 3.1 and Kling 3.0 were directly integrated into Arcads workflows in Q1, expanding the AI model options available for generating UGC-style video ads. The platform now offers 1,000+ AI actors on Pro plans with 29+ languages and regional accents, plus AI script generation with multiple hook variants.

Synthflow — (Updated) Primarily a voice AI platform rather than video, but crosses into creative workflows: Q1 highlights include a redesigned Agent Editor rolled out to all workspaces, Synthflow Chat (voice agents now also handle written conversation), Simulations 2.0 for automated regression testing, and a knowledge base recall improvement from 75% to 96%. Multi-Agent Systems with Subflows now support full GPT-5.1 and GPT-5.2.

Google Vids — (Updated) February 2026 brought a meaningful limit expansion: project max length increased from 10 to 30 minutes; recording studio max likewise extended; imported clips now up to 95 minutes or 4 GB. Veo 3.1's enhanced features (native vertical video, 4K upscaling) also rolled out inside Google Vids for Workspace users.

Deep Live Cam — (Updated) Version 2.6 (February 10) added Virtual Camera support — the AI face swap can now be hooked up as a virtual camera source for streaming platforms and video calls. The previous v2.3 had already introduced HyperSwap (200% better face swap quality) and 4× faster face enhancement.

Facefusion — (Still Relevant) Open-source local face manipulation tool; most recent release is v3.4.1 (September 2025), no new release in Q1 2026. Still actively maintained and widely used; 2026 installation tutorials proliferating on YouTube.

Amazon Nova Reel — (Still Relevant) Nova Reel 1.1 (April 2025) remains current with no new version in Q1 2026. Supports multi-shot videos up to 2 minutes via single-prompt or per-shot manual control, 1280×720 at 24fps, with invisible C2PA watermarking. Enterprise access via Amazon Bedrock.

Scenario — (Updated) Continuous model additions throughout Q1; 500+ models across 50+ providers accessible in one workspace. A new Asset Rarity Variants workflow (March 13) auto-generates Rare, Epic, and Legendary versions of game assets with escalating visual effects from a single base design. Core differentiator remains custom LoRA training from your own art bible for consistent game asset generation across image, video, audio, and 3D.

MovieFlo — (Still Relevant) Long-form AI movie generator; God Mode update (December 2025) added a dual-buff asset library and template library. No major new feature release in Q1 2026; platform described as "early days" with physics issues still present in reviewer testing.

Keytake — (Still Relevant) AI video generator for lead generation built around web-research-driven video composition. No specific new feature announcements in Q1 2026; platform continues on its existing feature set.

LTX-2 / LTX 2.3 (Lightricks) — (New) LTX-2 launched January 6 as the first production-ready open-source model to combine native audio and video generation with 4K output (19B parameters, 4K at 50fps, up to 20 seconds). Runs on consumer NVIDIA RTX GPUs; free for commercial use under $10M ARR. An ElevenLabs audio-to-video integration followed on January 19. LTX 2.3 (early March) scaled to 22B parameters with portrait mode, synchronised audio in a single pass, and four checkpoint variants. Immediately integrated into ComfyUI, fal.ai, Replicate, and OpenArt.

PixVerse R1 — (New) Alibaba-backed PixVerse launched a real-time AI "world model" on January 13 allowing users to command characters in real time — directing emotions, dances, and poses as video streams. PixVerse v4.5 also released as a model update. 16 million monthly active users as of late 2025.

Helios (Peking University / ByteDance / Canva) — (New) Autoregressive diffusion model released in March 2026 under Apache 2.0. Generates up to 1,440 frames (~60 seconds at 24fps) at 19.5fps on a single H100 GPU with no KV-cache or quantization tricks. Supports text-to-video, image-to-video, and video-to-video via unified input representation. 14B parameters.

Music & Audio AI Tools

ElevenLabs — (Major Update) The busiest period in the company's history. Eleven v3 launched in alpha on March 6 with inline Audio Tags ([excited], [whispers], [gunshot]) for mid-script emotion and sound FX control, multi-speaker dialogue via the new Text-to-Dialogue API endpoint, 70+ language support, and a 68% drop in error rate (from 15.3% to 4.9%). v3 went Generally Available on March 14, with users preferring it 72% of the time over the alpha. Conversational AI 2.0 (also March 6) is an enterprise voice-agent platform with natural turn-taking, multilingual auto-detection, integrated RAG for knowledge bases, HIPAA compliance, and EU data residency. ElevenLabs also repositioned itself as an "Audio OS" encompassing TTS, voice cloning, ElevenMusic, sound effects, dubbing, and transcription — with image/video generation integrations added. Eleven v3 is offered at 80% off until end of June 2026.

Udio — (Updated) Operating in a transitional state following its October 2025 copyright settlement with Universal Music Group. During Jan–Mar 2026, downloads remain disabled; tracks can only be streamed and remixed within Udio's licensed ecosystem with fingerprinting and content filtering active. A new model trained exclusively on authorised UMG catalog is in development for Q2 2026, with planned features including artist-voice/style creation (opt-in), AI remixing, mashups, multi-language vocals, and emotional tone nuance.

Suno — (Updated) Suno Studio 1.2 released February 6, adding Warp Markers, Remove FX, Alternates, and Time Signature support. V5 (launched late 2025) remains the active model through the period, delivering studio-grade audio quality, 4-minute song support, and an editor-first approach for structural control. No V5.x or V6 confirmed for Q1 2026; Suno staff stated significant 2026 changes will be communicated in advance.

Play HT / PlayAI — (Shutdown / Discontinued) Meta acquired Play AI in July 2025 for approximately $45M; the API went offline July 26, 2025 and the full platform shut down December 31, 2025. The 9-person team joined Meta's Superintelligence Labs; the voice technology is being integrated into Meta AI, AI Characters, and Wearables. ~40,000 users displaced; ElevenLabs and Murf AI ran migration promotions.

Artlist AI Voiceover — (Updated) A major voiceover upgrade landed March 11: 13 new languages added (including Dutch, Japanese, Korean, Russian, and Turkish), new English accent variants (American, British, Australian, Indian), and an Adobe Premiere Pro extension that generates AI voiceovers directly inside the editing timeline. A new Artboard from Script feature generates curated music, sound effects, and footage recommendations based on a voiceover script. Artlist also hosted a New York event in the period unveiling its 2026 "AI Toolkit" roadmap.

Hume Octave (Hume AI) — (Updated) Octave 2 (launched October 2025, active throughout the period) brought 11 languages, 40% faster responses at under 200ms latency, voice conversion, and phoneme editing. In January 2026, Hume published research on its training approach — 200+ emotions, 400+ voice characteristics, 0% error rate on phone number reproduction with targeted fine-tuning. In March 10, Hume open-sourced TADA (Text-Acoustic Dual Alignment model): real-time factor of 0.09 (5× faster than comparable LLM-based TTS), zero hallucinations in testing, on-device deployment support, English and 7 languages.

SoundHound AI — (Updated) Nasdaq-listed (SOUN) voice AI company posted 59% year-over-year revenue growth in Q4 2025, though the stock is down ~26% YTD in 2026. Amelia 7 platform expansion now handles multi-agent voice transactions (food ordering, hotel/flight bookings, parking payments, calendar management) across vehicles, TVs, and smart devices. Vision AI for vehicles was launched, combining in-vehicle cameras with speech recognition for landmark ID, sign translation, and hands-free troubleshooting.

Amazon Nova 2 Sonic — (Still Relevant) Nova 2 Sonic (launched December 2, 2025) is the current version for the period: a speech-to-speech real-time conversational AI model with polyglot voices that code-switch across languages mid-conversation, 1M token context window, and asynchronous tool calling. Supports Portuguese and Hindi beyond the core European languages; integrates with Amazon Connect, Twilio, Vonage, LiveKit, and Pipecat. Outperforms OpenAI gpt-realtime and Gemini 2.5 Flash on Big Bench Audio reasoning evaluation.

Google Lyria 3 — (New) Launched February 2026 in the Gemini app (all 18+ users in 8 languages). Creates 30-second tracks from text, photo, or video prompts with automatic lyric generation and controls for style, vocals, and tempo. Powers YouTube's Dream Track for Shorts. All outputs carry SynthID AI-content watermarking.

Mureka (Merika) — (New) Gaining traction in 2026 as a competitor to Suno and Udio, with music generation, lyric tools, model customisation, and a developer API for TTS and music generation. Cited by Curious Refuge in their 2026 AI music tools roundup as rivalling the major names across the category.

Sonauto — (New) Free, unlimited AI music generation platform geared toward gamers and short-form creators. v3 model entered public beta in February 2026, with a full release expected around April 2026. Notably accessible entry point for creators who need background music without licensing concerns.

Higgsfield Audio — (New) Launched March 2026 as part of Higgsfield's platform expansion. Consolidates ElevenLabs and Minimax audio models into a single interface; enables voice swapping and video-to-audio translation across 10 languages. Positioned as an entry point for creators already using Higgsfield's video tools who want audio handled in the same environment.

UI Design AI Tools

Figma — (Major Update) Figma pushed an enormous amount of product in Q1. Figma Make — prompt-to-prototype generation — rolled out to all paid tiers including Enterprise. The Figma MCP Server (March 2026) enables two-way workflows with dev tools: push rendered UI to canvas as editable frames, pull design context back into code, and connect GitHub Copilot users via VS Code. February's release wave added Modal components, Vectorize, Erase Object, a Cut tool, an expanded Slots component system (open beta), Glass effect improvements, and Claude Code-to-Figma integration. AI credit purchasing is now available on a pay-as-you-go basis for admins.

Google Stitch — (Major Update) Galileo AI was acquired by Google in mid-2025 and relaunched as Google Stitch inside Google Labs, powered by Gemini. The standalone Galileo product no longer exists. Stitch is free in beta: text-to-UI, image/sketch-to-UI, conversational refinement, Figma export with editable layers and auto-layout, and HTML/CSS/React/Tailwind code output. A Prototypes feature (multi-screen interactive flows) arrived with Gemini 3 integration in late 2025 and is active throughout this period. Standard generations run on Gemini 2.5 Flash; Experimental tier uses Gemini 2.5 Pro (350 + 50 generations/month free).

Webflow — (Major Update) Webflow's AI site builder was significantly rebuilt on 5 February 2026: multi-page sites (up to five pages) can now be generated from a single creation flow, built-in animations are included at generation, and Enterprise accounts gained access. Webflow also completed its GSAP acquisition and added Lottie animations support, a next-gen CMS (2× storage), deeper analytics with goal tracking, and an AI assistant that generates layouts, refines copy, and surfaces SEO/accessibility issues. Webflow Cloud is expanding toward AI-assisted full-stack app generation — blurring the boundary between website and web application.

Firebase Studio — (Major Update) Firebase Studio had a productive quarter. January brought Agent Skills — modular AI upgrades that reduce hallucinations and token costs via progressive disclosure, with five core Firebase Agent Skills. February doubled default max output tokens (8,192 → 32,768), added Gemini 3.0 Flash Preview and 3.0 Pro Preview models, upgraded Security Rules generation to Gemini 2.5 Pro, and introduced character-level (not just block-level) code edits in inline chat. Firestore Pipeline Operations now chain complex queries and aggregations without manual indexing.

Same.dev — (Updated) Same.dev clones any website with pixel-perfect accuracy from a URL, generating an editable React project — or builds from scratch via chat. The platform now runs on Claude-4.5-opus and has crossed 30 million projects built. Government, financial, and gambling sites are excluded from cloning. No major feature announcement in the window, but the milestone and model upgrade are noteworthy.

Uizard — (Still Relevant) Autodesigner 2.0 remains the current flagship: conversational UI generation combining ChatGPT-style iteration with live component editing, Screenshot Scanner (upload any app screenshot → editable UI), and Wireframe Scanner (sketch photo → digital wireframe). The Figma plugin continues to see adoption among UX practitioners in 2026. No new version shipped in this window; still a strong entry point for non-designers and early-stage founders. Pricing: Free (3 gen/month), Pro $12/month, Business $39/month.

Dora — (Still Relevant) Dora remains in Alpha 2.0 with free full access — prompt-to-site generation with 3D assets, scroll-triggered animations, Figma plugin, and responsive layouts. AI-generated 3D and AI-generated animations are listed as "in progress for 1.0" and have not yet shipped. No new release in this period. Worth watching for the 1.0 launch, but currently light on AI generation features for the category it's positioned in.

Attention Insight — (Updated) The Adobe Express Add-on v2 (released 23 December 2025, active throughout Q1 2026) brought predictive eye-tracking directly into Adobe Express workflows. New features include Attention Hotspots with percentage metrics and AI Recommendations for headings, CTAs, and main content — no manual input required, working as an overlay layer inside Express. The core platform continues to offer AI heatmaps claiming up to 96% accuracy against real eye-tracking data. No additional major release in the Jan–Mar window.

Operative.sh — (Still Relevant) Operative.sh provides browser-based agents for testing coding agent outputs — building, iterating, and deploying with AI that writes and tests apps as it goes. Limited public-facing release activity in this period; the GitHub organisation has two active repositories. One to watch if you're running agentic dev workflows that need automated UI verification.

Marketing AI Tools

Pencil — (Major Update) Three significant moves in Q1. Veo 3, Google's latest video generation model, is now live inside Pencil for video ad creation. The Editor Unification beta brings image generation and AI post-generation tools (Fill, Expand, Enhance) into the Social Ads Editor — create and edit without switching tabs. Pencil Pro is also now available on Google Cloud Marketplace with Gemini 1.5 Flash and Imagen 4 support. Used by four of the world's top ten advertisers via The Brandtech Group. Pricing: Core $14/month, Growth $55/month.

Jasper — (Major Update) Two major releases back-to-back. January 2026 shipped an Optimization Agent for SEO/GEO/AI-native content discovery (no Semrush account required), autonomous web and knowledge search, upgraded Nano Banana image generation, and OAuth for API. February added Jasper Grid (a new content execution system), Content Engineering certification, updated Optimization Agent scoring against Google's Search Quality Rater Guidelines, and MCP/API agent support for governance. Jasper's 2026 State of AI in Marketing report (1,400 marketers surveyed) found 91% of teams now use AI — up from 63% in 2025 — but only 41% can prove ROI, down from 49%. Scaling quality and governance are the dominant pain points.

Virallyst — (Still Relevant) Positioned as "Cursor for Marketing," Virallyst automatically generates social content each morning from live trend monitoring across YouTube, newsletters, and blogs — tailored to your brand voice without manual prompting. Includes AI visuals, scheduling, and auto-posting. $20/month. No product updates announced in this window, but the positioning is well-suited to solo operators and small creative agencies.

AiSDR — (Still Relevant) AiSDR published its 2026 State of the AI SDR Industry Report this quarter, finding AI-led outreach converts at 14.2% vs 3% for human SDRs when fully personalised. The platform covers email, LinkedIn, and SMS with 300+ contextual data sources. Pricing is usage-based (not per seat): Explore $900/month, Grow $2,500/month. A relevant benchmark for creative agencies and studios evaluating outbound tooling — though the entry price reflects enterprise positioning.

Keak — (Still Relevant) Keak's AI agent autonomously generates, tests, and deploys A/B test variants on websites — claiming a 73% win rate across 1.37M+ variants tested. AutoPilot mode continues to the next test automatically once a winner is found. Integrates with Shopify, Webflow, WordPress, Framer, and Squarespace. At 1.4M+ weekly users, it's quietly become the dominant AI CRO tool for no-code builders. $29/month.

Reachy — (Still Relevant) Desktop app (Windows) for LinkedIn outreach that respects platform rules through local, client-side operation. GPT-4o personalised messages, ICP lead scoring, signals-based search (people recruiting, attending events, reacting to posts), and a LinkedIn WarmUp system for safe scaling. A LinkedIn AI Bundle was teased in January 2026 but not yet released. $49/month for the Sales Agent tier.

Warmy — (Still Relevant) Email warmup platform actively addressing the 2026 privacy shift — Apple and Google privacy updates mean open rates "no longer tell the full story," and Warmy's content reflects that repositioning. Adeline AI handles warmup via a real inbox network with 30+ industry topics. No major feature launch in this period; the story here is that email deliverability tooling is navigating measurement collapse. $49/month starter (per inbox).

Happenstance — (Still Relevant) Natural-language network search across your professional connections — useful for creatives and agencies trying to find relevant contacts before pitches, events, or introductions. Crossed 200,000 users. A LinkedIn post in February 2026 raised questions about the data access model; worth reading their privacy policy if your network data is sensitive.

Gumloop — (New) AI automation platform that connects any LLM to internal tools without code. Described by marketing practitioners as "the most underrated AI tool" in 2026 roundups; MCP integration recently launched. Used internally by teams at Webflow, Instacart, and Shopify. Worth evaluating for content operations, data enrichment pipelines, and campaign automation.

Paradigm AI — (New) Described as a "horizontal Clay" — data enrichment at scale across marketing, sales, recruiting, and finance, pulling from social platforms and public data sources. A newer entrant gaining traction in performance marketing and research-heavy creative operations.

Pomelli — (New) Google Labs tool that ingests a website and generates on-brand social and ad content that matches the design system. Strong fit for DTC brands running performance ads who want creative that looks native rather than AI-generated. Currently in Google Labs.

No-Code Builders (Vibe Coding)

Lovable — (Major Update) Lovable 2.0 landed in January 2026 with two flagship modes: Agent Mode (autonomous execution — the agent plans, browses the web, searches the codebase, runs tests, and surfaces results without intervention) and Plan Mode (structured proposal for approval before building). Prompt Queuing lets you stack multiple prompts; the agent processes them in sequence. Native MCP integrations cover Notion, Linear, Jira, Confluence, and n8n. Lovable Cloud bundles database, auth, storage, and functions as a managed backend. ~8M users, $100M ARR milestone, $1.8B valuation.

Replit Agent — (Major Update) Agent 4 launched 13 March 2026 with a Design Canvas for generating UI mockups before writing any code. Plan Mode presents an estimated step-count before execution. Pro and Enterprise users can run parallel agent tasks simultaneously; real-time collaboration lets multiple users watch agent actions live. New connectors include BigQuery, Linear, Slack, and Notion. Mobile app deployment (iOS via Expo Go), in-app payments via RevenueCat, and a dedicated video processing stack were also added. $240M revenue in 2025; targeting $1B by end of 2026.

V0 by Vercel — (Major Update) V0 rebranded from v0.dev to v0.app in January 2026, alongside a new sandbox-based full-stack runtime. GitHub Import lets V0 agents edit and extend real codebases. A Git Panel (January) creates branches and PRs directly from chat. February added Snowflake and AWS database integrations for full-stack data apps and switched from credits to transparent token-based billing. Python service generation is now available alongside frontend components. 6M+ users, ~$42M ARR estimate, ~25% month-over-month growth.

Windsurf — (Major Update) Wave 13 (January) introduced Multi-Cascade Panes (multiple AI agents in parallel panes), Git Worktrees per agent, and SKILL.md project-level skill definitions. Wave 14 (February) added Arena Mode (side-by-side model comparison with preference voting), Plan Mode, the in-house SWE-1.5 model, 20× faster context loading for large repos, and Devin integration following Cognition AI's ~$250M acquisition of Windsurf in early 2026.

GitHub Copilot — (Major Update) Agent mode went generally available for all JetBrains IDEs on 11 March 2026. Custom Agents, Sub-Agents, Agent Hooks (automations triggered by GitHub events), Auto-Approve for MCP tool calls, a /memory slash command for codebase persistence, and a Thinking Panel for reasoning traces all shipped in this window. VS Code got parallel subagent execution and Codex agent access in January. Auto Model Selection (GA) now routes each task to the best available model automatically.

Warp — (Major Update) The Oz Agent (announced February 2026) is a cloud-based collaborative AI agent for terminal workflows: multi-repo changes in a single session, full terminal and computer use for self-verification, and MCP integrations with Linear, Figma, Slack, and Sentry. A WARP.md config file sets repo-level agent instructions, compatible with CLAUDE.md and SKILL.md. Universal input accepts drag-and-dropped files, images, and URLs directly into the terminal prompt.

Junie by JetBrains — (Major Update) Junie CLI beta launched 9 March 2026 — a standalone coding agent that runs from terminal, inside any IDE, or in CI/CD pipelines. Fully LLM-agnostic: bring your own key for OpenAI, Anthropic, Google Gemini, or Grok. One-click migration from Claude Code, Codex CLI, or Gemini CLI. Supports project-level junie.md instruction files, custom agents, and MCP. Pricing: $10/month individual, $60/month enterprise.

JetBrains Air — (New) Brand-new agentic development environment from JetBrains, built on the abandoned Fleet IDE codebase, launched in public preview on 9 March 2026. Run Codex, Claude Agent, Gemini CLI, and Junie simultaneously inside one environment via Multi-Agent Concurrency. The new Agent Client Protocol (ACP) handles agent-to-IDE communication; each agent task runs in an isolated Git worktree, Docker container, or cloud environment. macOS only at launch. Direct competitor to Google Antigravity and Cursor.

Google Antigravity — (New) Google's free AI-first IDE, released as a VS Code fork in November 2025 and actively updated throughout Q1 2026. Mission Control is a multi-agent orchestration dashboard for assigning tasks to multiple agents and monitoring progress. Model support includes Claude Opus 4.6, GPT-OSS-120B, Gemini 3.1 Pro, and others via a selector. No subscription required during preview — Google account sign-in only. Serious competition for Cursor and Windsurf in the AI IDE space.

Emergent — (Major Update) Full-stack natural-language app builder (web, mobile, landing pages, internal tools) that hit $100M ARR in February 2026. 6M+ users, 7M+ apps built, $100M total funding, Series B valued at ~$300M. Notably, 70% of users have no coding experience and 80–90% of new projects are mobile apps rather than web. A useful benchmark for the no-code market: the ceiling has moved considerably higher than early Bubble/Glide comparisons suggested.

Bolt.new — (Still Relevant) StackBlitz's browser-based full-stack app generator continues active development with ongoing model integrations and UI improvements throughout this period. No single flagship release to highlight, but it remains a strong competitor in instant-deploy full-stack generation alongside Lovable and Replit.

Base44 — (New) New entrant in vibe coding, emerging February 2026 with a simplified UX aimed at non-developers. Gaining traction in creator and solopreneur communities; rated favourably in community comparisons. Limited public documentation at time of writing — one to watch.

Local Models and Tools to Run Them

Tools

Ollama — (Major Update) A substantial January 2026 update. The new ollama launch command sets up Claude Code, OpenCode, or Codex CLI in one command, configuring local or cloud models automatically. Cloud Models (preview) let you run oversized models (70B+) on Ollama's own datacenter hardware via an API-compatible interface. GPT-OSS-20B and GPT-OSS-120B are now available via ollama pull. Context defaults are now set based on VRAM: <24 GB → 4K, 24–48 GB → 32K, ≥48 GB → 262K. Native image generation on macOS and refreshed desktop apps for macOS and Windows also shipped. Current version: v0.18.0.

Open Web UI — (Major Update) v0.8.0 (13 February 2026) was described by the project as its "largest release ever" — 30,000+ lines added across 300+ commits. New: OpenResponses (OpenAI-compatible API endpoint), Analytics Dashboard (token usage by model/user/time), Skills System, Message Queuing, Prompt Version Control, Action UI (HTML/iframe rich responses), 13× faster SCIM provisioning, and 34% faster authentication. Subsequent releases added Anthropic direct access (v0.8.4), Open Terminal with full file browser and script execution (v0.8.6), SQLite browser, Mermaid rendering, Jupyter previews (v0.8.9), and MariaDB Vector backend (v0.8.10, latest as of March 2026).

ComfyUI — (Major Update) Released v0.8 through v0.17 across January–March 2026. Key milestones: LTXV 2 model and Kling Omni (v0.8.0), native text generation nodes for Gemma3 and Qwen3 locally (v0.15.0), ElevenLabs API nodes for inline voiceover generation in image workflows (v0.15.0), Kling 3.0 Motion Control and xAI model nodes (v0.16.x), and a Painter node for embedded inpainting/masking (v0.17.0). Changelog. ComfyUI is now the de facto standard for local diffusion workflows and has expanded well beyond image generation.

Chatbox — (Updated) Three releases this period. v1.18.3 (13 January) added conversation search and auto-enabled web search. v1.19.0 (11 February) was the significant one: automatic context compression (summarises earlier conversation when nearing the limit), token usage percentage display, context length error detection, and a migration to AI SDK v6. v1.19.1 (27 February) added Nano Banana 2 image generation support.

Pinokio — (Still Relevant) One-click browser-based launcher for open-source AI apps (ComfyUI, Automatic1111, Open Web UI, etc.) — the fastest way to get local AI running without terminal setup. Current stable release is v3.9.0. Note: there is no in-app auto-updater; you must download new versions from the website manually.

Automatic1111 — (Shutdown / Discontinued) No meaningful update since v1.10 in 2024. GitHub discussions reflect broad acknowledgment that the project has stalled. If you're still running A1111, migrate to SD.Next / Vladmandic (actively maintained fork with Flux support) or Forge (most popular direct fork, adds Flux.1 and Chroma models with optimised memory management), or switch to ComfyUI for node-based workflows.

Models

GPT-OSS (OpenAI) — (Updated) Released August 2025 under Apache 2.0 — OpenAI's first open-weight models since GPT-2. GPT-OSS-120B fits on a single 80 GB H100 SXM; GPT-OSS-20B fits on a 16 GB edge device (Mac, consumer GPU). Both are reasoning models with selectable compute effort (low/medium/high). In January–March 2026, OpenAI released gpt-oss-safeguard variants — safety and PII classification models, also Apache 2.0. Available via ollama pull gpt-oss-120b and HuggingFace.

Llama 4 (Meta) — (Still Relevant) Scout (109B total / 17B active, 10M context, single H100) and Maverick (400B total / 17B active, 1M context, multi-GPU) remain the active open-weight releases throughout this period. Both are multimodal (text + image). Llama 4 Behemoth (2T parameters) is still in training with no release date. EU users still cannot download models directly from Meta due to an ongoing access restriction — a significant practical issue for European creatives and developers.

DeepSeek V3 — (Major Update) DeepSeek V3's context window expanded from 128K to 1M tokens on 11 February 2026. DeepSeek V4 was widely anticipated through this period (slipping from mid-February to March 10 with no official release as of March 17). A January training paper on manifold-constrained scaling was described by analysts as a methodological breakthrough. Still one of the most capable open-weight models available, and running locally at a fraction of the cost of API alternatives.

Mistral Small 4 — (New) Released 15 March 2026: 119B total parameters, 6B active (MoE with 128 experts, 4 active per token), 256K context window, open-weight on HuggingFace. A unified checkpoint handling instruction-following, reasoning, multimodal inputs, and coding without requiring specialist models. Also new this period: Voxtral (4 February) — a real-time speech-to-text translation model (4B params) designed for on-device, multi-language, streaming deployment.

Qwen 3.5 (Alibaba) — (New) Released 16 February 2026: 397B total / 17B active (MoE). Designed explicitly for the agentic era — strong tool use, multi-step planning, long-horizon tasks, and visual GUI interaction (can read and act on screenshots). 60% cheaper to operate than Qwen 2.5; 8× faster throughput. Available on HuggingFace and major inference providers.

Kimi K2.5 (Moonshot AI) — (New) Released February 2026: 1 trillion total parameters, 32B active (MoE with 384 experts), 256K context, native multimodal (text + image via MoonViT). Two inference modes — Thinking (chain-of-thought) and Instant (direct response). Open-source on HuggingFace. Agentic-first design optimised for tool use, web navigation, and multi-step planning.

Phi-4-reasoning-vision (Microsoft) — (New) Released 4 March 2026 under MIT licence. 15B parameters, multimodal (vision tokens injected mid-transformer via a "mid-fusion" architecture). Strong on maths, science, and UI grounding — it can read and reason about screenshots of interfaces, making it practically useful for design tooling and accessibility workflows. Outperforms larger models on multimodal reasoning benchmarks.

NVIDIA Nemotron 3 Super — (New) Released 11 March 2026: 120B total / 12B active, hybrid Mamba+Transformer MoE. 5× faster inference than comparable dense models. Unusually, NVIDIA released weights, training datasets, and training recipes — not just model weights. Part of NVIDIA's stated $26B open-source AI commitment. Adopted by Perplexity, Palantir, and Siemens at launch.

Ethical Models

Bria AI — (Major Update) Won two top awards at the 2026 Hollywood Professional Association Tech Retreat (17 March 2026). Bria's models are trained exclusively on 30+ licensed partner datasets — no unlicensed internet data — with a patented attribution system that traces every generated asset back to its source material. An initiative with Hollywood studios is developing a GenAI model trained on licensed studio libraries. For creatives working in commercial production, Bria remains the clearest answer to "can I actually use this in a client project?"

Other AI Tools

Cursor — (Major Update) Three significant moves in March 2026. Automations launched: always-on background agents triggered by Slack messages, Linear issues, GitHub PRs, PagerDuty alerts, or custom webhooks — each running in an isolated cloud sandbox and learning from past runs. The Cursor Marketplace arrived with 30+ plugins (Atlassian, Datadog, GitLab, Glean, Hugging Face, monday.com, PlanetScale) and support for private internal marketplaces. Release 2.6 added MCP Apps — interactive UIs rendered inside agent chats (charts, diagrams, whiteboards). JetBrains IDE integration shipped via Cursor ACP plugin in February. Business metrics: ~$2B ARR, ~$29.3B valuation.

Claude for Chrome — (Major Update) v1.0.62 (15 March 2026) added Workflow Recording (record once, replay automatically), Scheduled Tasks (time-triggered or event-triggered automations), Planning Mode (approve the entire plan before execution — not step-by-step), Claude Code integration (write-test-fix loops in-browser), Multi-Tab Management, and pre-trained knowledge of Slack, Gmail, Calendar, and GitHub UI patterns. Expanded from early access to all Claude Pro and Max subscribers in December 2025. Pro tier gets Haiku 4.5; Max gets full model access including Sonnet 4.5 and Opus 4.6.

Zapier MCP — (Major Update) AI Guardrails launched February 2026: detects 30+ types of PII (names, SSNs, credit card numbers), blocks Zaps containing sensitive data, detects prompt injection and jailbreak attempts, and adds sentiment scoring to AI-generated content before it sends. Zapier MCP now connects 8,000+ apps and 30,000+ actions to AI tools (Claude Desktop, ChatGPT, Cursor) via MCP. The go-to glue layer for enterprise AI agent stacks.

Manus AI — (Major Update) Meta acquired Manus for approximately $2B in December 2025. The Skills feature launched 28 January 2026: reusable SKILL.md files that teach agents specific workflows, shareable across teams via a Team Skill Library, with MCP tool compatibility. Platform now has 6M+ users; roadmap includes desktop integration, Meta glasses (Ray-Ban Meta), and native mobile development in beta for Max plan users.

Browser Use — (Still Relevant) Open-source browser agent framework (Apache 2.0) with 78,000+ GitHub stars and an 89.1% success rate on the WebVoyager benchmark. Multi-tab management, memory persistence across navigation, DOM-aware actions, and screenshot-based reasoning — works with any major LLM provider. No platform fees; pay only for LLM API costs. The most widely adopted open-source option for browser automation in AI agent stacks.

Stagehand — (Updated) v3.6.0 released 4 February 2026. Built on Chrome DevTools Protocol directly (no Playwright dependency), 44% faster than v2, model-agnostic agent mode, element caching, self-healing execution on DOM changes, and shadow DOM/iframe support. January–March additions: Claude 4 CUA mode support, GPT-5 API format compatibility, Gemini 2.5 Flash integration, and a new Evaluator class for automated success detection. 500,000+ weekly downloads.

Exa Search — (Major Update) Exa Deep relaunched 4 March 2026 with autonomous query expansion, an LLM reasoning layer that synthesises results, and field-level source grounding on structured outputs. Pricing update (3 March): search with contents now included free per request; Exa Deep reduced to $12/1,000 requests; new Exa Deep (Reasoning) tier at $15/1,000.

Gradio — (Major Update) Gradio 6.0 shipped January–February 2026 with native MCP Server support (any Gradio app can expose itself as an MCP tool), gr.HTML as a first-class layout element, custom templates and JS injection, typed server-side functions, and a push_to_hub CLI command for direct HuggingFace Spaces deployment. The go-to framework for rapid ML demo and API prototyping now integrates directly into agentic workflows.

GPT Realtime API — (Updated) gpt-realtime-1.5 released 23 February 2026: +5% on Big Bench Audio reasoning, +10% on alphanumeric transcription accuracy, +7% on instruction following, more reliable tool calling during voice sessions. Pricing unchanged. Migration required: gpt-4o-realtime-preview models are deprecated with shutdown scheduled for 7 May 2026.

Payman — (Still Relevant) Agentic payment infrastructure for AI-initiated transactions: agents send money under human-controlled policies (spending limits per agent, pre-approved payee lists). SOC 2 Type II + PCI DSS certified; Fifth Third Bank for USD custody; Stripe for processing. MCP Server connects Claude, Cursor, and other MCP-compatible agents to the payment APIs. Relevant for creative studios building automated supplier payments, AI freelancer payouts, or agent-driven purchasing.

Documenso — (Still Relevant) Open-source electronic signature platform (DocuSign alternative). API V2 shipped in late 2025/early 2026 with full multi-document envelope support and two-step direct upload. Self-hostable under AGPL; cloud-hosted option available. Active maintenance continues through Q1 2026. Relevant for creative businesses wanting document signing without SaaS lock-in.

Harvey AI — (Updated) Announced 110+ new regional legal data sources at Legalweek 2026 (March 2026), expanding from US-centric into EU, UK, and APAC jurisdictions. Leads in 7 of 11 legal AI use-case categories per the 2026 SKILLS industry survey including drafting, contract review, due diligence, and discovery. Relevant for creative agencies dealing with international licensing, co-production agreements, and IP negotiations.

Education AI Tools

NotebookLM — (Major Update) A busy Q1. Slide deck generation with PPTX export arrived February–March 2026: prompt-based revisions with controls for style, length, orientation (landscape/portrait/square), and detail level. Deep Research mode plans a research strategy, searches the web, and pulls 40–50+ sources (websites, GitHub repos, Reddit threads, papers) automatically. Google Drive search integration lets Deep Research surface relevant documents you already own. Image/OCR and CSV are now accepted as sources; chat history is now saved (previously ephemeral); prompt character limit expanded from 500 to 10,000. Moodle LTI integration is upcoming; Google Classroom integration allows notebooks to be created directly from class materials; NotebookLM notebooks can now be used as grounding sources inside Gemini.

Google Gemini Guided Learning — (Major Update) Significant education announcements at Bett UK 2026 (21 January 2026). Free full-length SAT practice tests built into Gemini (grounded in Princeton Review content) with immediate feedback and personalised study plans. A Gemini-powered Writing Coach on Khan Academy for grades 7–12 guides structure, thesis, and revision without writing essays for students. Gemini can now draft assignments and summarise student progress in Google Classroom using real class context. Premium Gemini Workspace features are now free for all Google Workspace for Education editions. Oxford University partnership extends access to all students and faculty.

OpenAI Study Mode / ChatGPT — (Updated) Study Mode (launched July 2025) is now available on all plans globally — Free, Plus, Pro, Team, and Edu — across iOS, Android, and web. Deep Research improvements (10 February 2026) added focus on specific websites and trusted sources, a redesigned sidebar, fullscreen report view, and the ability to create and edit a research plan mid-run. GPT-5.4 Thinking shipped with improved deep web research and better context window management.

Claude for Education — (Updated) Anthropic published the AI Fluency Index on 23 February 2026 — a landmark education research report analysing 11 observable AI collaboration behaviours across 9,830 conversations. Key finding: 85.7% of conversations showed iteration and refinement behaviour; iterative users are 5.6× more likely to question Claude's reasoning. The report is practically useful for educators designing AI-integrated curricula. Claude is also reported as producing "the biggest single jump" in educator productivity for generating polished Word documents from a single prompt.

OpenAI Academy — (New) Free Spring 2026 programme of livestreamed workshops launched ~5 March 2026: ChatGPT for Teachers, Codex for Software Engineers, ChatGPT for Resumes and Interviews, K–12 Applications, Managing District-Wide Adoption, and more. OpenAI Certifications are also in pilot — using ETS psychometric design and Credly digital badges, with a stated goal of certifying millions of workers. ChatGPT Foundations for Teachers is live via Coursera and moving into ChatGPT itself in early 2026.

NVIDIA AI Courses — (Updated) NVIDIA launched a new Agentic AI certification (NCP-AAI) in 2026 — testing ability to build and govern advanced multi-agent systems with scalability and ethical safeguards ($200, 2 hours). The full 2026 certification portfolio refresh covers Data Science, Physical AI, OpenUSD, and AI Infrastructure, with a webinar announcing updates scheduled 30 April 2026. GTC 2026 (16–19 March, San Jose) is running this week with focus areas on physical AI, agentic AI, and AI factories.

TurboLearn (Turbo AI) — (Still Relevant) AI lecture-to-notes platform with 4M+ student users. Core offering transforms lectures, PDFs, and recordings into structured notes, flashcards, and quizzes instantly. No major feature release announced in this specific window; the platform continues iterating on its core workflow. Still one of the most widely adopted AI study tools in higher education.

Globe Explorer — (Still Relevant) Free, no-registration visual knowledge exploration tool that generates structured, Wikipedia-style pages with visuals for any topic via LLMs. Now offers Default, Pro, and Turbo Pro modes. No major new features announced in this window, but it remains a useful entry point for creatives doing unfamiliar research — particularly for world-building, conceptual development, and cross-discipline research.

Maven — (Updated) Several AI bootcamps active in Q1 2026: AI Engineer Complete Bootcamp ($199), AI Builder Bootcamp for Product People, and the AI Learning Accelerator Program (launching March 2026) — a strategic/executive programme for mid-career professionals with twice-weekly live sessions focusing on AI strategy, governance, and deployment rather than coding. AI Engineering Buildcamp: From RAG to Agents starts 13 April (taught by Alexey Grigorev).

Microsoft Generative AI / Elevate for Educators — (Major Update) Microsoft announced Elevate for Educators at Bett 2026 (15 January 2026): AI Skills Navigator (self-paced courses + live sessions + AI-powered simulations in 13+ languages), a new free AI in Special Education course, the Microsoft Elevate Educator Credential (with ISTE + ASCD), and the Microsoft Learning Zone app with on-device Copilot+ PC features. Four new AI certifications went GA in February 2026 including AI Business Professional, AI Transformation Leader, and Agentic AI Business Solutions Architect. The Generative AI for Beginners course (18 lessons) remains active on Microsoft Learn.

Roadmap.sh AI Tutor — (Updated) A new AI Agents roadmap published at roadmap.sh/ai-agents: "Learn to design, build and ship AI agents in 2026." The 27 February 2026 changelog added Claude Code, Vibe Coding, and AI in Guides updates — making this a useful current-state reference for self-directed learners entering the agentic tooling space.

Worth Checking Out

Halo by Brilliant Labs — (New) AI smart glasses began shipping Q1 2026 with "Noa" — a private, conversational AI agent with long-term memory and Vibe Mode for natural-language building. A partnership announced 5 March 2026 with Neuphonic and TheStage AI moves AI processing onto the glasses and paired smartphone (on-device inference). The broader AI glasses market in 2026 now includes 10+ products: Meta Ray-Ban with display, HTC VIVE Eagle, INMO Air 3 AR, Even Realities G2, RayNeo X3 Pro, and VITURE Luma Ultra. AI wearable computing is no longer a prototype story.

GeoSpy — (Major Update) GeoSpy AI can now pinpoint the exact location of most social media photos at metre-level accuracy — "the era of anonymous backgrounds is ending" went viral in March 2026. Used by law enforcement for suspect location. For creatives and content teams: understand that AI can now reverse-engineer location from any background detail in your imagery. Relevant for location scouting, privacy-conscious content creation, and OSINT.

Figure 03 / Helix 02 — (Major Update) Figure AI unveiled Helix 02 on 27 January 2026: a single neural system controlling a full humanoid body from pixel input, combining System 0 (whole-body motion prior), System 1 (sensor-to-action), and System 2 (semantic reasoning). Figure 03 on Helix 02 autonomously unloaded and reloaded a dishwasher in 61 sequential actions without resets, using palm-mounted cameras and fingertip tactile sensing. March 2026 showed 8 new autonomous cleaning behaviours including coordinated tool use (spray bottle + towel), bimanual manipulation, and in-hand reorientation. The physical AI frontier is moving faster than most production timelines.

Devin 2.2 — (Major Update) Devin 2.2 launched 24 February 2026: 3× faster startup, Full Desktop Testing (end-to-end via computer use for any Linux desktop app), AskDevin expanded to Ask + Plan modes, Devin Fast Mode (~2× faster at 4× ACU per session), Skills support (reusable instructions from codebase), and smoother Slack and Linear integrations. Devin usage doubled every two months throughout 2025 (~65× growth over 13 months). Now supporting engineering fleets at global enterprises. Post-Cognition/Windsurf acquisition, Devin's async agent capabilities are being merged into Windsurf.

Mistral OCR 3 — (New) Mistral OCR 3 released January 2026: 74% win rate over OCR 2 on forms, handwriting, and complex tables. Priced at $2/1,000 pages ($1 with Batch API). Outputs Markdown with HTML table reconstruction; the Document AI Playground in Mistral Studio gives a drag-and-drop parsing interface. Also new: context biasing and diarization added to the Audio Transcriptions API (27 January 2026).

Terra Security — (New) Terra Portal launched 10 March 2026: agentic desktop app for professional penetration testers with two modes — Ambient (fully autonomous recon, code review, test generation) and Copilot (human-in-the-loop for exploitation steps). CI/CD integration for continuous security testing on code push; auto-generated compliance reports in SOC 2 and ISO 27001 formats. Beta testing reduced discovery-to-fix timelines from ~3 months to hours. Relevant for creative studios handling sensitive client data or building client-facing platforms.

Think Pieces and Resources

UNESCO Re|Shaping Policies for Creativity 2026 — (February 2026) The most authoritative global snapshot of AI's impact on creative economies. 267 pages covering 120+ countries and 4,000 policies. Key data point: music creators are projected to lose 24% of revenue by 2028; audiovisual creators face 21% revenue decline as AI-generated content expands. Policy responses are lagging market reality by a significant margin. Essential reading for anyone working on advocacy, contracts, or pricing strategy in the creative industries.

Jasper 2026 State of AI in Marketing — (January 2026) Survey of 1,400 marketing professionals. 91% of teams now use AI (up from 63% in 2025). Only 41% can prove ROI — down from 49% last year, suggesting adoption has outpaced measurement. Scaling quality and governance are the dominant pain points, not access or cost. Free to download; a useful benchmark for positioning creative AI services to marketing clients.

Anthropic AI Fluency Index — (February 2026) Research analysing 11 observable AI collaboration behaviours across 9,830 conversations. The headline: iterative users — those who refine and push back — get measurably better outputs and are 5.6× more likely to question Claude's reasoning. The practical takeaway for creatives: the quality of your AI collaboration is more about your conversational behaviour than your prompt engineering skills.

Morrison Foerster AI Copyright Trends 2026 — (February 2026) Describes 2026 as peak AI copyright litigation year. Courts are developing a judicial consensus that training general-purpose AI is "highly transformative," but significant disagreements remain on other issues. The US Supreme Court declined to hear the Thaler case on 2 March 2026, leaving the human authorship requirement for copyright intact. For creatives using AI commercially: document your human creative involvement. For European practitioners, the EU AI Act's full application on 2 August 2026 is the compliance deadline for high-risk systems.

NVIDIA State of AI 2026 — (March 2026) Enterprise survey findings: 86% of AI budgets are increasing in 2026; data analytics (62%) and generative AI (61%) are the top workloads; agentic AI is rising sharply; lack of AI experts is the #1 adoption challenge. Open source is driving enterprise AI strategy. A useful calibration for creative technologists positioning themselves in the enterprise market.

Deloitte State of AI in the Enterprise 2026 — Worker access to AI rose 50% in 2025. Companies with ≥40% of AI projects in production are set to double in six months. Only 1-in-5 companies has mature governance for autonomous AI agents. 58% report at least limited use of physical AI — projected to reach 80% in two years. The governance gap is an opportunity for creative studios and consultancies.

Bill Gurley on the AI Bubble (CNBC, March 16, 2026) — "The AI wave is real and has made a lot of people rich quickly — but that leads to bubbles. A reset is coming." Salesforce and ServiceNow are down ~25% YTD; the iShares Software ETF is down ~20%. Benchmark's Gurley isn't calling the end of AI, but he's recommending investors identify target prices for distressed SaaS stocks. The counter-signal: big tech AI capex is running at ~$700B in 2026. The gap between infrastructure investment and software sector performance is a pattern worth watching.

Sound on Sound AI Music Tech 2026 — Survey of musicians, producers, and engineers. Two-thirds believe EDM and mainstream pop are most vulnerable to AI displacement. Only 9% expect full automation; 21% expect major automation with human oversight. Production timelines in the commercial sector have reduced by 60%+ since 2024. Contextual music-to-video sync is now viable. A grounded read on the music industry's actual position — not the hype version.

January–March 2026 Trends and Market Highlights

Agentic AI is no longer a pilot programme — it's in production. Gartner forecasts that 40% of enterprise applications will embed task-specific AI agents by the end of 2026, up from under 5% in 2025. What's changed in this quarter is that the infrastructure has caught up with the ambition: MCP as an open protocol (now contributed to the Linux Foundation), Agent-to-Agent (A2A) communication specs from Google, Anthropic's open governance model, and tools like Zapier MCP connecting 8,000+ apps mean agents can now operate across fragmented enterprise toolstacks without custom integration work. For creative professionals, this translates directly: your project management tools, design environments, and communication platforms are becoming orchestratable from a single agent layer. The caution — flagged by both Gartner and Kore.ai — is that over 40% of agentic AI projects are predicted to be abandoned by 2027, typically because teams deployed high-autonomy agents in domains requiring constrained, auditable behaviour. Design before you deploy.

The open-source model landscape has fundamentally changed the local-model conversation. Mistral Small 4 (119B parameters with only 6B active per token), Qwen 3.5 (60% cheaper than its predecessor), Kimi K2.5 (1T total params, 32B active), NVIDIA Nemotron 3 Super (5× faster inference than comparable dense models), and — most significantly — OpenAI's own GPT-OSS models under Apache 2.0 have collectively removed the argument that local models are a quality compromise. Running a frontier-competitive model on a Mac or a consumer GPU was the domain of enthusiasts six months ago; it's now a practical choice for studios and independent practitioners who want to avoid per-query API costs and keep client data on-premise. For EU-based creatives: note that Meta's Llama 4 models remain inaccessible for direct download in the EU, making Mistral and Qwen particularly relevant locally-run alternatives.

The vibe coding category is consolidating around a smaller set of much more capable platforms. Lovable 2.0 (8M users, $1.8B valuation), Replit Agent 4 (Design Canvas, parallel tasks), V0's rebranded full-stack runtime, and Emergent's $100M ARR milestone signal that this category has moved past proof-of-concept. The significant insight from Emergent's data is that 70% of its users have no coding experience and 80–90% of new projects are mobile apps — the "non-developer building for their business" use case has overtaken the "developer using AI as a shortcut" use case. For creative studios, the implication is that internal tools, client portals, and bespoke creative apps are now within reach of design-led teams without an engineering budget. The Cognition AI acquisition of Windsurf (~$250M), Manus's acquisition by Meta (~$2B), and Play HT's absorption into Meta's Superintelligence Labs all point toward consolidation accelerating.

Copyright and creator compensation is reaching a pivotal inflection point. The US Supreme Court's refusal to hear Thaler v. Vidal on 2 March 2026 preserved the human authorship requirement for copyright — purely AI-generated works remain unprotectable, and this looks settled for the near term. Morrison Foerster has called 2026 "peak AI copyright litigation year." Meanwhile, UNESCO's Re|Shaping Policies for Creativity 2026 report quantifies the economic pressure: music creators face a projected 24% revenue decline by 2028, audiovisual creators 21%. Udio's settlement and licensed rebuild with Universal Music Group, and Warner Music's settlement with Suno, represent the music industry's pragmatic response — building licensed AI into the ecosystem rather than trying to litigate it out. The EU AI Act's full application on 2 August 2026 adds a compliance layer that European creative businesses cannot ignore: high-risk systems must have completed conformity assessments, CE marking, and EU database registration by that date, with fines up to €35M or 7% of global turnover. The UK's copyright and AI full report with policy options was due on 18 March 2026 — the creative content exchange marketplace pilot is already running.

February 2026 was the largest single startup funding month on record at $189B globally, with OpenAI's $110B raise (Amazon $50B, Nvidia $30B, SoftBank $30B) and Anthropic's $30B Series G at the top of the table. ElevenLabs raised $500M at an $11B valuation. The capital concentration at the foundation model layer is stark: seed-stage funding fell 11% year-on-year in February even as total funding hit records, suggesting the investor thesis has narrowed sharply to infrastructure and the handful of established platform players. For creative tool companies, this is both a validation — AI creative tooling is considered critical infrastructure — and a warning: the mid-tier SaaS layer (Salesforce down ~25% YTD, ServiceNow down ~25%) is under pressure from AI commoditisation of workflows those tools were built around. Bill Gurley's March 16 warning about an AI bubble "reset" is worth reading alongside the NVIDIA survey finding that 86% of enterprise AI budgets are still growing.

The physical AI and spatial computing layer is arriving faster than most practitioners anticipated. Figure 03 running Helix 02 autonomously completing 61-step household tasks, Apptronik raising $520M at $5.5B+ valuation, and World Labs raising $1B for 3D world models are not abstract future signals — they are commercially staged events on a 12–18 month deployment timeline. Spatial computing deal value exceeded $47B in 2025. For creative and UI/UX practitioners specifically, the question is no longer "will interfaces move beyond screens?" but "what design language and interaction paradigms work for embodied AI and spatial computing?" The fact that Phi-4-reasoning-vision can already read and reason about UI screenshots, and that Qwen 3.5 can perceive and interact with GUI elements, suggests that interface design is becoming a substrate for AI action rather than just human perception.

AI hallucination entered the scientific record in 2026. GPTZero found 100+ AI-hallucinated citations across 53 accepted NeurIPS 2025 papers — each passing three to five human peer reviewers — and 50+ confirmed hallucinations in ICLR 2026 submissions. This is not primarily a story about AI failure; it's a story about human over-reliance on AI-assisted research in time-pressured workflows. For creative professionals using AI to research markets, cite sources, or produce factual copy: verification is not optional. The 94% of Duke students who believe AI accuracy varies significantly by subject are correct, and the behaviour gap between users who iterate and question AI outputs versus those who accept them (as measured by Anthropic's AI Fluency Index) produces meaningfully different work quality. The skill being separated from the tool is critical judgment — and that is genuinely a craft skill rather than a technical one.

If you spot any missing links or updates, please DM or comment!