New AI releases we tracked, newest first, each linked to its official source. Rankings last updated . RSS feed.
Tencent's new flagship LLM adds visual understanding and computer use (command line and desktop apps) for the first time, with stronger long-context and long-horizon agent work. Live in Yuanbao, CodeBuddy and WorkBuddy, with the API on Tencent Cloud TokenHub and OpenRouter at about $0.90/M input and $2.70/M output tokens; Tencent says open weights are coming in the next few days.
Pick GitHub Copilot in JetBrains Air's agent picker to delegate coding tasks on your existing Copilot subscription, locally in your JetBrains IDE or as cloud tasks through Air Teams that run in isolated containers and bill your Copilot plan instead of JetBrains AI credits.
Describe a plugin in plain language and DeepSeek Harness builds and installs it, extending the open-source agent app's tools, skills and interface (DeepSeek's demos include a design workspace, a video-editing workbench and a 3D LLM visualizer). Harness is free, open source and in public preview.
Long conversations and agent loops on the Claude Messages API can now keep the conversation on Anthropic's servers, so each later request sends only the new messages and gets a faster first token. In beta on the Claude API, AWS, Google Cloud and Microsoft Foundry; a thread expires 5 minutes after its last request.
Meituan's new native multimodal model for long-horizon agent work: 1.6 trillion total parameters (about 48 billion active per token), built on LongCat-2.0 with stronger visual understanding and agentic coding, including visual coding, design and 3D tasks. Try it at longcat.ai; an API launched alongside it.
Anthropic renamed Claude Max 5x to Max ($100/month) and Max 20x to Ultra ($200/month), with no price change. Ultra gives 10x Pro usage each week and drops the 5-hour limit for verified accounts, starting with most existing subscribers; weekly limits are unchanged.
Google moved the Gemini API's Computer Use tool, which lets a model operate user interfaces for agent tasks, to general availability across the Interactions and GenerateContent APIs.
Mistral released the weights of its improved 1-trillion-parameter multimodal flagship, Le Chonk (public preview since early October), with gains in agentic coding and defensive cybersecurity and three reasoning modes. Free to use and commercialize under a modified MIT license (Model-as-a-Service providers need a separate commercial license); API $0.70/M input and $2.10/M output tokens.
Pika's open-weights video model, post-trained from MiniMax H3: Pika says it makes a 15-second 768p clip with synced audio in as little as 7 seconds of compute, takes up to 12 references, exports up to 5 audio stems and finishes at up to 4K. In Pika's API now; on pika.art from November 2.
Generate, preview and edit documents, PDFs, slide decks, spreadsheets and code right inside a Gemini chat: type edits full screen, or highlight text to ask questions or request changes. PDFs work on web and mobile, other formats on web only; available to all Gemini users.
Canva Research's vision-language model for design editing: restyle a design, change every page of a deck in one prompt, or remix assets for new formats, with everything staying fully editable. Powering design edits in Canva AI for Canva Pro subscribers now, wider availability coming soon.
ElevenLabs' newest text-to-sound-effects model follows prompts with several sounds more reliably, handles music and vocals better and sounds cleaner than SFX v2. Available in ElevenCreative, the mobile app and ElevenAPI (limited access, not in every region yet).
An AI agent that builds editable, node-based 3D workflows: chain generation, texturing, retopology, UV, PBR, rigging and export, then save, reuse and batch-run them across assets. In beta and rolling out gradually, with early access for Max and Team users; free users get 24 hours of access.
Google's new frontier model, its strongest yet at cybersecurity and long-horizon coding and agent work; rolling out first to a subset of cyber defenders through Google's Fairwind Program, with developer and consumer access to follow. Introductory API price $2/M input and $10/M output tokens (later $4 and $20).
GitHub's Copilot app adds Projects (public preview), where you give Copilot a goal plus repos, designs and notes and it runs as many agents as needed in the background, and computer use (public preview on macOS and Windows) to operate desktop apps after asking your approval.
Instead of picking a model, pick an effort level (Standard, High or Max) and Copilot routes each task to one model or orchestrates several (cascade or critique). Generally available on all Copilot plans in VS Code, the Copilot app, Copilot CLI and JetBrains; billed at each model's standard rate, with a 30% discount on paid plans for a limited time.
DeepMind's new world model that generates interactive, real-time simulations with native audio (music, ambient sound and dialogue) you can steer in plain language; research preview for trusted testers now, coming to Google AI Ultra in Project Genie and early-access Gemini API partners, with a waitlist.
Liquid AI's new open-weight family of six small (2B-class) text, vision, audio, omni and decision models built to run on-device and be fine-tuned on a single GPU; the text model has a 256K context and fits in about 1 GB at 4-bit. On Hugging Face and the Liquid AI API.
Microsoft's image models are now generally available in Microsoft Foundry, with multi-reference editing (up to five images), Bing web grounding and output up to 1.5K; 2.6 starts at $5/M text-input and $38/M image-output tokens, Flash at $1.75 and $19.
Google's open-source, zero-dependency gemini-api command-line tool moved to beta: it lets coding agents like Claude Code, Codex and Cursor generate text, images, speech and video, analyze files and run Gemini managed agents from the shell.
With the 27.2 software updates, Apple's rebuilt Siri AI (in beta) now works in French, Japanese, Korean, Portuguese and Spanish as well as English, on Apple Intelligence devices; still unavailable in the EU on iPhone, iPad and Watch.
OpenAI's new fast, lower-cost model for visual reasoning, computer use, coding and everyday writing, in ChatGPT Work, Codex and the API (gpt-6.1-terra) at $1/M input and $5/M output tokens, with a 1M-token context.
The best new AI tools, ranked, in your inbox every week. Free.
Get the loot →Free. Unsubscribe anytime.