The AI Wire ●

Moonshot AI released Kimi K3.1 Pro on Friday, a post-trained version of Kimi K3 with a 1 million-token context window, live now on Kimi.ai, the Kimi apps, Kimi Work, Kimi Code and the Kimi API. Moonshot calls it its most capable 3T-class open-weight model to date: 2.8 trillion total parameters with 104 billion active per token, text and image input, priced at $2 per million input tokens and $10 per million output — a third below K3 — with a Highspeed variant running two to three times faster at double the price.

Artificial Analysis scores it 55 on its Intelligence Index, fourth overall and an 11-point jump over K3, behind only Claude Opus 5.5, Claude Sonnet 5.5 and Muse Wonder Full. In Moonshot's own table it scores 65.7% on Terminal-Bench 4.0 against 66.4% for Opus 5.5, though almost every coding score was run inside Anthropic's Claude Code harness and the figures are not independently checked.

The full weights are promised by November 4; if they land as promised, K3.1 Pro would become the highest-scoring open-weights model on the Index. Moonshot also says it deliberately removed hacking-related data from training ahead of the open-weight release.

Reuters filled in the cyber curb on October 24: Moonshot said it removed cyber-related data not just from the training corpus but from the reinforcement-learning environments too, while keeping general coding skills. 'We aim not to release open weights that would extend the absolute frontier of cyber capability,' the company said; on its internal vulnerability-research test the final model scored 81.0%, against 90.5% for a checkpoint without the change (vendor-reported). The restraint lands a week after China proposed export controls that would ban exports of AI-driven malware-generation technology.

Update, November 4: the weights are out, two days ahead of the promise. Moonshot published the full model on Hugging Face on November 2 as moonshotai/Kimi-K3.1-Pro — 96 files, about 1.56 TB, not gated — under the Kimi K3.1-Pro License, word for word the K3 terms: free use, but a model-as-a-service business above $20 million in revenue over 12 months needs a separate agreement with Moonshot, and products above 100 million monthly users or $20 million in monthly revenue must display the model's name. Artificial Analysis was still listing the model as proprietary on November 4, two days after the upload.

Meta's personal AI agent Muse launched in Japan on October 23, its first market outside North America since its September debut in the US and Canada. Alexandr Wang, Meta's chief AI officer, said the agent was available to download from app stores on the day, weeks after Meta's Japanese launch page had said local availability was still undecided.

Muse is available in Japan on iOS, Android and the muse.ai website, with many basic functions free and subscription plans for heavier use, though the Japan announcement does not specify local prices or free-tier limits. The agent moves beyond answering questions to carrying out tasks such as sending emails and booking travel through services users connect.

The expansion follows a fast North American start: Meta said on its October 21 earnings call that more than 10 million people in the US and Canada had set up Muse in its first six weeks. Japan is the first test of whether that uptake travels, and of how Muse handles language, local services and pricing outside its home market.

Google released Gemini Notebook as a Connected App in the Gemini app on Friday, deepening an integration that previously lived in the app's side panel. Typing @Gemini Notebook in a chat lets users manage and query their notebooks in natural language, and Gemini can also detect on its own when a prompt calls for it.

From inside a chat, users can create, edit or delete a notebook, add, update or delete sources, list notebooks and their sources, save a chat into a notebook, and ask questions answered from a notebook's contents. The feature is on by default for users globally, except in the European Economic Area, where it is off by default.

It is available to all Google Workspace customers, Workspace Individual subscribers and users with personal Google accounts. Admins control it through the Gemini and Gemini Notebook access settings for their organization.

Anthropic declared the general availability of its unified Claude experience for Enterprise customers on October 23, extending the combined chat-and-Cowork surface introduced for Pro and Max subscribers in September. A conversation now starts as an ordinary chat on web, desktop or mobile, and when a request needs code execution or file work, Claude escalates into an Anthropic-managed cloud workspace in the same thread — work that continues even after the user closes a laptop, and can be resumed on another device.

Each cloud session runs in an isolated virtual machine. For Enterprise, network access from that environment is disabled by default, and connector credentials stay outside the sandbox, while existing chats, projects, connectors, skills and plugins carry over into the combined experience. Existing sign-in and compliance controls continue to apply.

Administrators can enable the unified experience early or keep it disabled; it will be turned on by default for eligible Enterprise organizations on November 16. A new Efficient mode restricts the combined experience to Sonnet and Haiku models, giving administrators another way to control spending alongside organization, group and individual limits. Pricing remains usage-based.

OpenAI made its Decisions API generally available on Thursday, taking the GPT-6 Luna-powered endpoint out of public beta. The API returns bounded, typed answers — a probability, a selection, or a rubric score — that applications use to classify content, route requests, and prioritize work.

The price is $0.10 per million input tokens, the same headline rate as GPT-6 Luna itself. OpenAI says focusing Luna's intelligence on questions with bounded answers delivers those answers up to 10 times faster than asking Luna through the Responses API, across text, image, and structured inputs.

The endpoint first surfaced at DevDay alongside the Agents API, then opened to public beta on October 6. General availability makes it a standard part of OpenAI's developer platform, aimed at the many production calls that need a decision rather than a paragraph.

Canva released its first image generation model on Thursday. Canva Craft, built by the company's own research team, is live for all users in the Dream Lab tab and available to developers through the Canva API, TechCrunch reported. It is aimed squarely at the same-category models from Google, OpenAI, ByteDance and Meta that Canva previously rented for its platform.

Craft works in two stages: a planning stage first works out the spatial relationships of elements, their relative scale, the visual hierarchy and any text, and a rendering stage then generates the image at up to 2K resolution across some 30 aspect ratios. The differentiator is the training data — human-designed templates from Canva's own library, including posters, social posts, presentations, infographics, menus and other typography-rich layouts, rather than the photograph-heavy datasets behind most rivals. That makes for legible text and precise edits that leave the rest of an image untouched, and generated images can be converted into layered, editable designs with Canva's Magic tools. Researchers from Leonardo.ai, which Canva acquired in 2024, contributed to the model.

Preliminary blind comparisons put Craft fourth on DesignArena's image-generation leaderboard, which would make Canva the second-highest-ranked image lab behind OpenAI. Cheaper Craftfast and Craft-ultrafast variants are planned, as is image generation inside third-party chatbots such as ChatGPT and Gemini through Canva's MCP connector.

Anthropic has made Claude Models on Google Cloud (Anthropic managed) generally available, giving Google Cloud customers a second way to buy Claude alongside the existing Google-managed offering. Both routes use Google Cloud credentials, IAM and Marketplace billing, and spending on the Anthropic-managed service counts toward existing Google Cloud commitments — but Anthropic operates this service itself, sets its usage limits and acts as the data processor under its own commercial terms.

The practical difference is features. The Anthropic-managed route carries the full Claude Platform from day one, including document uploads, web search and code execution, with new models and platform features arriving the same day Anthropic releases them; Claude Managed Agents remains in beta. The launch line-up is Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, and Opus 5.5 fast mode is generally available through the service at $8 per million input tokens and $40 per million output tokens.

Google's own comparison keeps the Google-managed option as the better fit for customers who need Google as data processor, strict regional residency, or native integrations such as Model Armor and Agent Engine, and it does not claim the same VPC Service Controls and customer-managed encryption keys for the Anthropic-managed route. Anthropic also warns customers with existing discounts to talk to their account executives before starting, because discounts cannot be applied retroactively to usage on the new service.

Recursive Superintelligence released its first model on Wednesday: Flint, an open-weight coding model designed to run on a laptop rather than a data center. Flint is a mixture-of-experts model with 39.8 billion total parameters but only about 2.5 billion active per token, fits on a 48 GB Apple Silicon Mac or a single 48 GB Nvidia GPU (4-bit weights take about 23 GB), and is free on Hugging Face under Apache 2.0. The model card lists a 262,144-token context window; Recursive's blog cites up to 512k.

Recursive says Flint generates more than 125 tokens per second on an M4 Pro MacBook Pro and more than 150 on Nvidia RTX Spark — up to twice comparable models. On its own charts it tops other MoE models under 50B on agentic coding, scoring 74.2% on SWE-bench Verified and 95.1% on IFEval, though it trails badly on Terminal-Bench 2.1 at 54.9% against 73.0% for a dense Qwen model. The company concedes Flint may need more guidance than large closed coding assistants.

The bigger claim is how Flint was made: a small team built it from scratch in under four months, leaning on Eureka Machine, Recursive's automated research system, for nearly every part of the stack. Recursive, led by former Salesforce chief scientist Richard Socher, emerged from stealth in May with $650 million in funding from backers including GV, Nvidia and AMD Ventures, and pitches Flint as a first public test of recursive self-improvement.

Google Maps users on Apple CarPlay in the US can now tap a Gemini icon on the car display and ask questions in natural language while driving. It is the first time Google has made a voice assistant available inside its CarPlay app; Google Assistant was previously only available when navigating from the phone itself.

Drivers can ask what the weather will be at their destination on arrival, or ask multi-step questions such as where to stop for sightseeing and where to eat or get fuel along the way. Google says the feature is rolling out now in the US, with a global rollout planned.

Anthropic released an updated Claude Security on Wednesday, in public beta for Claude Enterprise customers. Instead of one-off scans, the tool now monitors connected repositories continuously: after a deep baseline scan, it rescans code as it changes, based on each change's risk and complexity.

Findings are now tracked individually. When Claude Security detects the code change that fixes a vulnerability, it marks the finding fixed automatically, and can reopen it if the issue resurfaces. The update also adds per-project threat models, multi-repository projects of up to eight repositories — so it can catch issues that live between services — and a new API, alongside a companion Claude Code plugin beta for local, session-specific scans.

Anthropic says that since Claude Security entered public beta in April, thousands of organizations have used it to find over a million vulnerabilities. The product is powered by Anthropic's full model family, including Claude Mythos, with no Cyber Verification Program access required, and eligible organizations get $25,000 in free usage credits per month for two months, on usage through December 31, 2026.

Anthropic has launched Claude API version 2026-10-21, the first formal API version since 2023-06-01, according to its ClaudeDevs account. The headline change for developers: the Messages API no longer accepts the temperature, top_p and top_k sampling parameters.

The new version also returns a 400 error for documents passed in an assistant turn or citations in a user turn, and replaces some deprecated Admin and Enterprise Analytics API fields. New major versions of the Anthropic SDK and the ant CLI default to 2026-10-21 and ship with migration guides.

The old 2023-06-01 version is not deprecated, so existing integrations keep working. Anthropic says it will release API versions more frequently from here, so developers can absorb incompatible changes before new models require them.

OpenAI is rolling out Locked Chats in ChatGPT, letting users move selected conversations out of their regular history, pinned chats and search into a protected space. Locked chats open with a four-digit account PIN on web and desktop, or a PIN, Face ID, Touch ID or device unlock on supported mobile devices.

The October 20 release notes describe a gradual rollout across Free, Go, Plus and Pro, with Business, Enterprise and Education tiers also included, so the option may not appear in every account at once. New information remembered inside a locked chat stays separate from regular conversations, though memories saved before a chat was locked are not automatically removed.

There are sharp edges. Locking a chat does not change OpenAI's retention or training settings for it, and if you forget the PIN and reset it, the protected chats and their related content are permanently deleted. The feature also does not cover temporary sessions, archived threads, group chats or Codex conversations.

Nvidia's GEAR robotics lab open-sourced Flora on October 20: three models that turn robot and human manipulation video into the labels robot AI needs. Flora-TempSeg (9B) splits a video into action intervals, Flora-Captioner (27B) writes one short instruction for each interval, and Flora-Rewarder (27B) scores clips for outcome and process rewards, including against rubrics a user writes. All three are fine-tuned from Alibaba's Qwen3.5 models, with weights on Hugging Face under an Nvidia non-commercial research license.

Nvidia says Flora is the data engine used to train its GR00T humanoid models, labeling up to 200,000 hours of video a day across 512 GB300 chips. Its headline comparison is cost: about $0.72 to label an hour of video, against about $88.49 for GPT-6 Astra at maximum effort — roughly 125 times cheaper — with a 57.1 percent overall annotation score for Flora against 53.7 percent for Astra in Nvidia's own 50-episode evaluation.

Both figures need their caveats. The cost gap rests on Nvidia's modeled GPU pricing rather than measured production bills, the quality evaluation was run by Nvidia itself, and the results vary by task: Flora led on human videos and action boundaries, while Astra led on robot videos, caption accuracy and rubric scoring.

Mistral AI released Voxtral 3 on October 20, a family of speech-to-text and text-to-speech models plus a voice agent SDK that combines listening, reasoning, tool calling and speaking in one real-time streaming pipeline. The models are available now through Mistral's API and can also be self-deployed on a company's own infrastructure.

On transcription, Voxtral Small Transcribe 3 supports 50 languages with speaker diarization and context biasing at $0.0040 per minute, while a smaller Mini Transcribe 3 starts at $0.0035 per minute in batch and a realtime version built for agents costs $0.008 per minute. For speech output, Voxtral Speech 3 covers more than 35 languages and 50 dialects, adjusting emotion and pacing, at $13 per million characters.

A Pro Voice feature adapts a speaker's voice and style across languages from 30 minutes of audio. Mistral says the transcription models outperform Whisper, Scribe v2 and Gemini Transcribe on its tests — company-reported claims, but the pricing puts Voxtral 3 squarely in the running for call-center and service-desk voice agents.

Grok Bot gained another batch of abilities on October 20, in a "What's new in Grok Bot" walkthrough posted on X. The agent can now read and reply to your X Chat direct messages, respond when you tag @bot in an X thread, and be set as the default digital assistant on Android phones.

The DM and tagging features need the X plugin installed and configured, and Palmer said limits are in place so the Bot cannot spam people. Voice calls also now keep the previous call and its full transcript if you call back within 30 minutes, alongside new chat cards, email aliases, connectors and team invites.

The Android move lands a day after Grok Bot gained one-press access through the iPhone's Action Button, which starts a voice chat with the Bot. Together the two put SpaceXAI's always-on agent one press away on both major phone platforms.