中文 ↗RSS ↗

THE NEWSROOM / AI 新闻

AI News · Page 13

Source-linked briefings on models, agents, products and practical AI.

YYSI publication dates appear on the left; original announcement dates appear under each brief.

Search and categories filter this page. Use the page links below for earlier news.

SOURCE

Product updates MiniMax

MiniMax Code launches Mini App runtime for persistent agent workspaces

MiniMax announced Mini App, a new extension format for its MiniMax Code desktop app that packages a frontend and Node.js backend as an installable plugin. According to the company, Mini Apps keep agent work on a live page rather than scattered across conversations, and can call and be called by the Agent. The release includes a runtime, a /miniapp-creator flow, and official Mini Apps in the plugin marketplace.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Practical guides Anthropic

Anthropic Opens Call for Life Sciences Guest Investigators

Anthropic announced an open call for principal investigators and research groups at academic institutions, nonprofits, and government labs to apply for 8-12 week guest investigator projects using Claude. The company says it will run its models and harnesses at no cost, collaborate on publishing outputs, and open-source accompanying data and code. Applications are open for roughly 14 days, with selections notified about two weeks after closing.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Product updates Anthropic

Anthropic-managed Claude Platform now generally available on Google Cloud

Anthropic says its Anthropic-managed Claude offering on Google Cloud is now generally available, giving customers the full Claude Platform feature set through Google Cloud IAM credentials, billing and commitment drawdown. The vendor claims new features and betas ship the same day as on the native platform. Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5 are listed as available, with US and Global endpoints.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Models & research Anthropic

Anthropic Launches Claude Haiku 5.5

Anthropic announced the release of Claude Haiku 5.5, which the company describes as its most economical, fastest and most capable small model to date. According to the release notes, it is designed for high-volume and cost-sensitive tasks. The announcement directs readers to a separate blog post for further details. No pricing, benchmarks or availability dates beyond the October 7, 2026 release note are provided.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Agents & tools Anthropic

Anthropic Opens Claude Duet Research Preview for Pro and Max Plans

Anthropic has opened a research preview of Claude Duet for Pro and Max plan users, according to its release notes dated October 21, 2026. The company describes Claude Duet as a new app in which Claude acts as a proactive agent. Anthropic points readers to a separate blog post introducing Claude Duet for further details. The release notes do not specify availability beyond Pro and Max plans.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Product updates Mistral

Mistral and NVIDIA partner on AI for industrial engineering

Mistral announced a collaboration with NVIDIA to build AI-native engineering design workflows, combining Mistral Large 4 with NVIDIA CUDA-X, Warp and Omniverse libraries for faster simulation and closed-loop design. The companies also plan public benchmarks and open datasets for physical and agentic AI in CAE and EDA, targeting aerospace, automotive and semiconductor use cases. Mistral says Mistral Large 4 has about 1 trillion total parameters and weights due late October.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Product updates Anthropic

Anthropic Releases API Version 2026-10-21 With Breaking Changes

Anthropic published API version 2026-10-21, which the vendor says is available on the Claude API, Claude Platform on AWS, Claude Models on Google Cloud, Microsoft Foundry, and on Amazon Bedrock for Claude 4.7 and later models. The release removes the temperature, top_p, and top_k request parameters, returning 400 errors when sent, and changes several Admin API and Claude Enterprise Analytics API response fields.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Product updates Anthropic

Anthropic adds HIPAA BAA coverage for local Claude Code and Cowork

Anthropic published documentation stating that Enterprise plan organizations can apply a HIPAA setting that brings local Claude Code and local Cowork sessions under a Business Associate Agreement. The setting is available only to organizations with an Anthropic account team, and only the primary owner can apply it. Cloud sessions, IDE extensions, connectors, MCP servers, and gateway or cloud-provider connections are excluded from BAA coverage.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Product updates Google

Gemini with Personal Intelligence rolls out to Android Auto and Google built-in cars

Google says Gemini is now available globally on Android Auto and cars with Google built-in, and it is rolling out Personal Intelligence starting today. With user permission, the feature connects information from Workspace apps, Photos, Maps, YouTube and Search to give tailored in-car answers, such as flight details or route-based restaurant picks. Google says it is off by default and can be disabled anytime in Gemini settings.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Product updates Google

Chrome expands Gemini-powered scam detection to all Enhanced Protection users

Google says Chrome has moved its Gemini-based scam detection from on-device processing to a more powerful server-side model, now available to all users who enable Enhanced Protection. According to the company, the model analyzes page content such as the brand represented, page intent and a scam-likelihood score, which Safe Browsing combines with domain reputation and other signals. Google also describes new suspicious-site warnings and clipboard analysis targeting ClickFix attacks.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Product updates Google

Google launches Gemini business hub and brand tools for small businesses

Google announced a new business homepage in the Gemini app aimed at small business owners, letting them manage branding, finances and customer insights in one place. The company says users can connect Google tools like Business Profile, Ads and Workspace, plus third-party apps such as Canva and Wix. Brand visuals can be set by uploading a site, brand kit, fonts or colors, with image generation handled by the Nano Banana 2.1 model. Google says the features are rolling out globally over the coming days.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Product updates Google

Google expands Financial Services Verification to US and Canada

Google announced it is bringing Financial Services Verification to the United States and Canada, adding a credential check on top of identity verification for certain financial ads. Advertisers must show authorization from relevant regulators or recognized registries before running those ads. Google says the program already covers about 2 billion users across 42 countries, and it plans a phased rollout with country-specific advertiser guidance ahead of early 2027.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Product updates Google

Google Wallet on Android adds My Number Card digital ID with selective attribute sharing

Google announced that Android users in Japan can now register their My Number Card in Google Wallet, starting today. The new attribute-certification feature lets users tap or scan a supported phone to prove identity, age or address without presenting the physical card, sharing only the fields they choose. Google says only minimal data is stored on-device, with PIN and biometric locks and remote data erasure, built on open standards such as ISO 18013-5.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Models & research MiniMax

MiniMax Releases M3.1 Flash With Faster Inference and Lower Cost

MiniMax has launched MiniMax M3.1 Flash, a next-generation Flash model it says improves agentic tasks, coding, computer use and cybersecurity. The company claims online inference is nearly three times faster and usage costs are 70% lower than its predecessor, with flat pricing of $0.15 per million input tokens and $0.60 per million output tokens. It is available via the MiniMax API, M Plan and MiniMax Code, with weights to follow later.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Models & research MiniMax

MiniMax Sets Retirement Schedule for M2.5 and M2.7 Models

MiniMax published a retirement schedule for its M2.5 and M2.7 model families. The vendor says the models will close to new users on October 19, 2026, and be fully retired on November 18, 2026, after which API requests will fail. M Plan and Token Plan traffic will be automatically routed to M3.1 Flash starting October 19, with existing model names kept for backward compatibility.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Product updates Anthropic

Anthropic commits shares to new Public Benefit Fund

Anthropic says it is committing a portion of its shares to the Anthropic Public Benefit Fund, a new wholly owned LLC. Subject to stockholder approval, 5% of shares will be committed now, with an additional 2% at the board's discretion for every 50% increase in company value above a set baseline. Anthropic says roughly 20% of shares will be dedicated to the public good when combined with prior pledges.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Models & research DeepSeek

DeepSeek upgrades Flash model to V4.1-Flash-1017

DeepSeek says it has upgraded its Flash model to DeepSeek-V4.1-Flash-1017, and that the official web and app models now use the new version. The vendor reports benchmark gains over V4.1 and says task runtime and output length were reduced. API calling method and pricing are unchanged, with access through the deepseek-flash model name. DeepSeek also opened its in-house search API.

AI summary · Official sourceOriginal source ↗Read brief →