67 notes
New releases
Shipped changes we could point at a public changelog or release note. Dates are the publisher's dates. Summaries stay close to what that page actually says.
ByteDance Seedream 5.5
ByteDance Seed introduced Seedream 5.5, an image model with more stable multi-turn editing across 100 rounds, native RGBA output including transparency, and more realistic images and text rendering. Evaluation scores on the post are vendor-reported.
Thinking Machines Inkling 2 Small
Thinking Machines introduced Inkling 2 Small, a mixture-of-experts model with 594 billion total parameters and 17 billion active, with image input and variable effort. It is available on Tinker with closed weights, and the lab says Apache 2.0 weights are planned after external safety testing. Benchmark scores on the post are vendor-reported.
Ant Ling-3.2-flash
Ant Ling's changelog for November 5, 2026 launched Ling-3.2-flash, a native multimodal model with 560 billion total parameters, 25 billion active, and a 1 million token context, taking image, text, and video input. It is available in the chat interface and through an API compatible with OpenAI and Anthropic protocols, and is free for the first 2 weeks.
Claude Personal Finance beta
Anthropic is rolling out Personal Finance in Claude in beta, with read-only bank and card connections through Plaid, on Pro, Max, and Ultra plans in the US. The article date is November 5, 2026.
Gemini Omni 1.2 Flash generally available
Google made Gemini Omni 1.2 Flash generally available on November 5, 2026, for generating and editing video up to 15 seconds, on the Gemini API and Google Cloud. Google says 720p video is $0.07 per second, 30% less than Omni 1.1; that price comparison is vendor-reported.
OpenAI GPT-6 Cyber
OpenAI introduced GPT-6 Cyber on November 5, 2026, a cybersecurity model initially available through the API and Codex to a select group of approved Daybreak Red customers. Benchmark comparisons on the page are vendor-reported.
Perplexity Teammates
Perplexity launched Teammates, agents for ongoing work that plan projects and bring results back for review. The blog is dated November 4, 2026. Perplexity's changelog says Teammates are available to Max subscribers on the web.
Tencent Hy4
Tencent announced Hy4, an open-source model with visual understanding and computer use. The announcement says the API is approximately $0.90 per million input tokens and $2.70 per million output tokens, and that weights and system cards are planned in the coming days. Performance comparisons on that page are vendor-reported.
Meituan LongCat-2.5
Meituan released LongCat-2.5, a native multimodal model for long-horizon agent tasks, with 1.6 trillion total parameters. The company says people can try it at longcat.ai and developers can call it through an API.
DiffuSpace Clover-35B-A3B-Preview
DiffuSpace released Clover-35B-A3B-Preview, an open-source Apache-licensed mixture-of-experts diffusion language model with 35 billion total parameters and 3 billion active, aimed at agentic coding and tool use. Benchmark and speed figures on the post are vendor-reported.
TBC mischief 1
The Biological Computing Co. launched TBC mischief 1, an image-to-video model built on Wan 2.2 and available on fal and OpenRouter. The company says it is 5 times faster than its base model at half a cent per second; those speed and price figures are vendor claims.
Mistral Large 4 open weights
Mistral released open weights for Mistral Large 4 (Le Chonk), a 1-trillion-parameter multimodal mixture-of-experts model, under the Mistral Ecosystem Open-Weight License, a modified MIT license that requires a separate commercial license for Model-as-a-Service businesses. The API price on the page is $0.70 per million input tokens and $2.10 per million output tokens. Coding and security rankings on the page are vendor-reported.
Pika H3 Flex
Pika launched Pika H3 Flex, an open-source video and audio model family post-trained on MiniMax H3, with weights on Hugging Face. Pika says it ranks number 1 among open-source text-to-video models; that ranking is a vendor claim.
Gemini app file editing
Google added file generation, preview, and editing in the Gemini app for documents, PDFs, slides, spreadsheets, and code. PDF preview and editing work on web and mobile; Google says the other formats are web only, and that this is available to all users where the Gemini app is available.
GitHub Copilot Fusion
GitHub Copilot Fusion is generally available in the Copilot model picker, with Standard, High, and Max effort and Single, Cascade, and Critique patterns, on all Copilot plans in VS Code, the Copilot app, Copilot CLI, and JetBrains. GitHub says usage is billed per model used, with a 30% discount for a limited time on paid plans, and a same-day note says local-model routing is an experimental preview. The changelog's coding-index comparison is a vendor claim.
GitHub Changelog (opens in a new tab)GitHub Copilot on AIToolStack
NVIDIA Nemotron 3.5 Transcribe
NVIDIA published Nemotron 3.5 Transcribe, an open-weight promptable speech model (30B-A4B, about 4.3 billion active parameters) for transcription, speaker attribution, translation, and timestamps, under the OpenMDW-1.1 license. Leaderboard and speed figures on the page are vendor-reported.
Gemini 4 Carbon, limited access
Google announced Gemini 4 Carbon, and it is not generally available. Access starts with a subset of cyber defenders through the Fairwind Program, with wider availability later, at an introductory $2 per million input tokens and $10 per million output tokens.
Particle-1
Particle launched Particle-1, a coding model post-trained on DeepSeek V4.1 Flash and available through the Particle API. Particle reports that it uses around 45% fewer tokens on average; that figure is vendor-reported.
iCloud Mail and Calendar in ChatGPT
OpenAI's English release notes for October 27, 2026 say iCloud users can connect Mail and Calendar in ChatGPT to find emails, catch up on an inbox, check a schedule, and update calendar events. It is available on all ChatGPT plans on the web, the desktop app, and iOS.
OpenAI release notes (opens in a new tab)ChatGPT on AIToolStack
Genie 4 research preview
Google DeepMind announced Genie 4, a world-simulation model, as a research preview for trusted testers. A waitlist is open for Google AI Ultra users in Project Genie before a later release to consumers and developers.
Liquid LFM3
Liquid AI announced LFM3, a family of open-weight models in the 2B parameter range covering text, vision, audio, omni, and decision models. The weights are on Hugging Face, and the models are also on the Liquid API.
Ling-3.1-flash-1019
inclusionAI published Ling-3.1-flash-1019 under the MIT license, with 560 billion total parameters, 25 billion active per token, and a 1 million token context window. Benchmark gains on the card are vendor-reported, and this date is the Hugging Face repository creation time because the card does not print one.
GPT-6.1 Terra
OpenAI introduced GPT-6.1 Terra for ChatGPT Work and Codex, and in the API as gpt-6.1-terra at $1 per million input tokens and $5 per million output tokens. The English release notes date the rollout October 26, 2026, include Plus, Pro, Business, Enterprise, and Edu, and say Terra is available in Work and Codex, not Chat.
OpenAI (opens in a new tab)ChatGPT on AIToolStackCodex on AIToolStack
Nanbeige4.3-33B
Nanbeige LLM Lab published Nanbeige4.3-33B, open weights under the Apache 2.0 license with a context of up to 512K tokens. Benchmark scores on the card are vendor-reported. The card does not print a calendar date; this date is the Hugging Face repository creation time.
NVIDIA SkillSpector v2.13.1
NVIDIA published SkillSpector v2.13.1, an open-source security scanner for AI agent skills under the Apache 2.0 license.
Kimi K3.1 Pro and Highspeed on Vercel AI Gateway
Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed are available on Vercel AI Gateway. Vercel describes Highspeed as a faster version for interactive use; K3.1 Pro keeps a 1M-token context window and image input, and the changelog says Moonshot plans to release the weights by November 4.
Canva Craft
Canva released Craft, its first image model, in Canva and through the public API. It generates images at up to 2K resolution across more than 30 aspect ratios and can render text in 55 languages.
OpenAI Decisions API generally available
OpenAI made the Decisions API generally available. It is built with GPT-6 Luna, and OpenAI reports it is up to 10 times faster than GPT-6 Luna through the Responses API, which is a vendor-reported comparison, at a stated $0.10 per million input tokens and $0.01 per million cached input tokens.
Google Cloud Agent Compute Engine announced
Google Cloud announced Agent Compute Engine, a hardware-isolated runtime for agents. The post says it will be available in preview in the coming months, so this note is an announcement, not a launch.
Mistral Voxtral 3
Mistral released Voxtral 3, a voice-agent SDK with speech models in more than 50 languages, available through Mistral's API or self-deployment. The article datePublished is October 22, 2026.
Claude Duet research preview
Claude Duet is a research preview, available by waitlist for US Pro and Max subscribers as a standalone app on iOS and Android and on claude.ai. For this preview it is included in Pro and Max and uses the same usage limits as the rest of Claude.
Gemini app Customize hub
Google added a Customize hub in the Gemini app for skills and connected apps, including more than 40 external services. It is rolling out globally on web, mobile, and desktop for all user tiers, and the gradual rollout started on October 13, 2026.
Google Workspace Updates (opens in a new tab)Gemini on AIToolStack
OpenAI models coming to Google Cloud
OpenAI announced that its models and Codex are coming to Google Cloud Model Garden, with general availability planned in the coming weeks. Microsoft remains OpenAI's primary cloud partner.
Nemotron 3.5 Embed, Rerank, and Parse
NVIDIA published open Nemotron 3.5 retrieval models: Embed 8B BF16, Embed 1B BF16, Embed 1B NVFP4, Rerank 8B BF16, and Parse at 631 million parameters. The post says they are commercially available.
Recursive Flint v1.0 on OpenRouter
OpenRouter lists Recursive Flint v1.0 as recursive/flint-v1.0:free, a free model with a 262K context window and about 2.5 billion active parameters out of 40 billion. The listing date is October 21, 2026.
Carbon-X on Hugging Face
Hugging Face published Carbon-X, a 14.8-billion-parameter model trained on DNA sequences and text, under the Apache 2.0 license. The model card does not print a publish date; this date is the Hugging Face repository creation time.
Krea 3 out of beta
Krea's October 20, 2026 changelog says Krea 3 is out of beta.
Eleven Realtime v1 Mini and v1 Max
ElevenLabs' October 19, 2026 changelog adds Eleven Realtime v1 Mini and v1 Max for ElevenAgents voice agents, rolling out gradually. Each model is a complete voice-agent configuration, and the changelog says an OpenAI Realtime API client can connect by changing the endpoint and session configuration.
ElevenLabs changelog (opens in a new tab)ElevenLabs on AIToolStack
Ling-3.0-flash-Sante
inclusionAI published Ling-3.0-flash-Sante, a healthcare reasoning model under the MIT license, with 124 billion total parameters and 5.1 billion active per token. The card says it is a research model and not a medical device. The card does not print a calendar date; this date is the Hugging Face repository creation time.
Gemini API CLI experimental preview
Google released an experimental preview of the Gemini API CLI (gemini-api), a zero-dependency binary for macOS, Linux, and Windows that calls the Gemini API from the terminal.
Inception Mercury 3
Mercury 3 is available on the Inception API and through OpenRouter, Baseten, and Vercel, with a 1 million token context, at $0.20 per million input tokens and $0.70 per million output tokens. Inception reports 1,068 tokens per second and 4–8× faster running; those speed figures are vendor-reported.
MiniMax M3.1 Flash
MiniMax M3.1 Flash is available through the MiniMax API, the M Plan, and MiniMax Code. Model weights are public on Hugging Face under the MiniMax Community License (https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash). MiniMax reports nearly 3× faster inference and 70% lower usage cost, which is vendor-reported, at a listed $0.15 per million input tokens, $0.60 per million output tokens, and $0.015 per million cached input tokens.
Meta Muse Glimmer 30B
Meta published Muse Glimmer 30B, an open-source 30-billion-parameter model for on-device agents that takes text and images and returns text, quantized to about 4-bit. The page releases it under the Apache 2.0 license.
DeepSeek-V4.1-Flash-1017
DeepSeek-V4.1-Flash-1017 is live on the DeepSeek website, app, and API, with weights on Hugging Face and ModelScope. Search and Fetch are listed at $0.75 per 1,000 calls, and the page says billing for those calls through DeepSeek Harness, Claude Code, and Codex starts October 31, 2026 at 16:00 UTC.
Kling 4.0 global launch
Kling AI announced the global launch of Kling 4.0, with native 30-second generation and up to 15 reference assets and 10 keyframe inputs. The release says 10-bit HDR output and developer API access are coming soon.
Kling AI via GlobeNewswire (opens in a new tab)Kling AI on AIToolStack
Canvases from Cursor cloud agents
Cloud agents can now build interactive canvases as part of their work. Share a finished canvas with your team by sending its link.
Updates to GPT-6 Pro in ChatGPT
GPT-6 Pro in the chat tab now offers faster responses, improved use of plugins and tools, and multiple agents that can work on different parts of a task in parallel. You can follow their progress in the conversation and open completed tasks to see what each agent did. This update is rolling out to eligible Pro users in ChatGPT.
ChatGPT release notes (opens in a new tab)ChatGPT on AIToolStack
Gemini 4 Argon in Perplexity
Gemini 4 Argon is now available in Perplexity for Pro and Max subscribers.
Pika plugin for ChatGPT
The official Pika plugin lets ChatGPT use Pika's creation apps to make and edit images, video, music, sound effects, and speech without leaving ChatGPT.
Mirelo SFX 2.0 in Runway
Mirelo SFX 2.0 is on Runway. It generates sound effects, foley, and ambience from a video while keeping the dialogue.
Perplexity Decider V1.2 on Vercel AI Gateway
Perplexity Decider V1.2 27B and Decider Flash V1.2 4B are now available on Vercel AI Gateway. Both are multimodal decision models for classifying content, routing requests, and scoring text, structured records, and images.
Vercel changelog (opens in a new tab)Perplexity on AIToolStack
REA skill in Perplexity Computer
REA (Reverse Engineer Anything) is now a skill in Perplexity Computer. Ask Computer to reverse-engineer an app and it inspects how the app works without its source code, then builds a working version.
Perplexity Computer (opens in a new tab)Perplexity on AIToolStack
Krea 3 launch
Krea 3 launched on krea.ai, with up to 6 reference images per edit, and is also available through fal, Hedra, Together AI, Nous Research, OpenRouter, and Replicate. Krea reports edits in under 8 seconds at native 2K, which is vendor-reported, at $0.06 per image at 2K and $0.03 at 1K.
Meta Muse Spark Open
Meta released Muse Spark Open weights, an 816-billion-parameter model with 42 billion active parameters and a 1 million token context. The weights are under the Muse Model License.
Mistral Timestral
Mistral released Timestral, its first time-series model family, in four sizes (5M, 28M, 123M, and 209M parameters) with a context of 32,768 steps and a modified MIT license. Mistral reports that the largest model ranks first on GIFT-Eval, TIME, and fev-bench; that ranking is vendor-reported.
Nemotron 3.5 Super VL and Bolt
NVIDIA released Nemotron 3.5 Super VL, an open 120-billion-parameter model, and Nemotron 3.5 Bolt 74B-A9B, both available now, with Bolt E4B coming October 23. The post's 5× faster claim is vendor-reported.
GitHub in Slack and Teams, multiple repositories
@GitHub in Slack and Microsoft Teams can work across multiple repositories in one conversation, including issues, pull requests, and bug investigations. The changelog also notes mention-anywhere, easier repository switching, and Slack thread context.
GitHub Changelog (opens in a new tab)GitHub Copilot on AIToolStack
GPT-Live-1.1 in ChatGPT Voice
OpenAI is rolling out GPT-Live-1.1, and a smaller Quip variant, in ChatGPT Voice, with more natural accents across languages and additional voices. Available voices vary by region and account.
OpenAI release notes (opens in a new tab)ChatGPT on AIToolStack
Figma Specs open beta
Figma Specs is available on web, desktop, and mobile, rolling out in open beta to paid plans.
Google Deep Review public launch
Google Research publicly launched Deep Review for technical documents. The post says each account gets two complimentary reviews per week, a Gemini API key raises the limit, and the first inputs are PDFs and plain text.
Gemini 4 Argon in GitHub Copilot
Gemini 4 Argon is generally available in GitHub Copilot for complex coding, agent, and reasoning tasks. Copilot Pro+, Max, Business, and Enterprise can select it. Rollout is gradual.
GitHub Changelog (opens in a new tab)GitHub Copilot on AIToolStack
Gemini 4 Argon in Google Antigravity
Google is rolling out Gemini 4 Argon in the Antigravity desktop app and CLI for Google AI Pro and Ultra subscribers. Businesses can use it through the Gemini platform API and in Antigravity on eligible Gemini Enterprise subscriptions.
Z.ai GLM-Legal
Z.ai published GLM-Legal, a model for agentic legal work. API access is open as a beta, and AutoClaw support is described as coming later.
Composer predictions in Codex
Codex can suggest the next message from the current thread. You accept it with Tab, then review or edit it. Accepting a suggestion does not send it.
OpenAI release notes (opens in a new tab)Codex on AIToolStack
Claude Haiku 5.5
Anthropic launched Claude Haiku 5.5, its fastest small model, for high-volume, cost-sensitive tasks.
Claude release notes (opens in a new tab)Claude on AIToolStack
Claude Sonnet 5.5 in Cursor
Cursor's web release notes for the week of October 5, 2026 list Claude Sonnet 5.5 as a new model in Cursor.
Cursor web release notes (opens in a new tab)Cursor on AIToolStack
Seedance 2.5 Draft mode in Runway
Runway's agent can generate with Seedance 2.5 Draft mode: fast 480p drafts, then a post-generation enhancement. Draft mode can be toggled on or off.