AI Weekly · Week 46, 2026 · 11.09 — 11.15
STARHOT
Google DeepMind released multiple cutting-edge models and tools, including Gemini 4 Carbon, AlphaProof Code, and Gemini Robotics 2.5, and opened multi-agent workflows in Antigravity. Anthropic launched Claude Fable 5.2 and Claude Mythos 5.2 on AWS. Tencent's Huanyuan Hy4 was released, supporting native visual and computer operations for the first time.

Anthropic launched Claude Fable 5.2 and Claude Mythos 5.2 on AWS
Anthropic Launches Claude Fable 5.2 and Claude Mythos 5.2 on AWS, Fable 5.2 is its most powerful general model for coding and scientific fields, supporting unsupervised operation for days; Mythos 5.2 is provided through a validation program and performs more reliably in cybersecurity and life science fields.
Model launches & updates
In this sectionGoogle DeepMind released multiple cutting-edge models and tools, Anthropic launched a new version of Claude, and Tencent's Huanyuan Hy4 was released, supporting native visual and computer operations.

Google DeepMind releases the Gemini 4 Carbon cutting-edge model
Google DeepMind has opened Gemini 4 Carbon to developers, enterprises, scientists, and consumers, claiming it has reached cutting-edge levels in network security capabilities and long-range intelligent agent workflows.

Google DeepMind releases AlphaProof Code, a four-week verification of the Signal encryption core
Google DeepMind releases AlphaProof Code, an AI-powered mathematical proof software that meets security specifications, and completes the verification work usually requiring several years on the Signal encryption protocol in just four weeks.

MiniMax releases open-source image generation model H3-Image
MiniMax has released an open-source weight image generation model based on H3, H3-Image, which supports text-to-image, reference image generation, and image editing, capable of generating 4096 × 4096 4K images within 10 seconds, and supports up to 16 reference images.

Google DeepMind releases Gemini Robotics 2.5 vision-language-action model
Google DeepMind released Gemini Robotics 2.5, claiming it is the most capable vision-language-action (VLA) model to date, with a 30% to 100% increase in success rate on 200 tasks compared to previous baselines, and 20% to 30% faster task completion speed.

Tencent Hunyuan Hy4 released: 770B total parameters, 49B activation parameters, first supports native vision and computer operation
Tencent Mixuan Hy4 is officially released, with a total of 770B parameters and 49B activation parameters. It introduces native visual capabilities and computer operation support for the first time, while also upgrading command compliance, long text understanding, and long-range Agent capabilities. The model has been launched on products such as Tencent Cloud TokenHub, WorkBuddy/CodeBuddy, and Yuanbao. API access has been provided to platforms like OpenRouter, Hermes, and OpenCode, with weights to be open-sourced in the near future.
Product launches & updates
Google Opens /teamwork Multi-Agent Workflow in Antigravity
Google has opened the /teamwork multi-agent workflow in Antigravity to external researchers and developers, which can be used in conjunction with Gemini 4 Argon, Carbon, and Hydrogen; /boost and Deep Review are also available.
Google expands the Antigravity multi-agent system, launches the Teamwork system, and verifies multiple mathematical proofs.
Google announces expansion of its Antigravity multi-agent system, forming the Teamwork system with multiple professional agents driven by Gemini, capable of verifying complex mathematical proofs and optimizing software system performance.

Google launches Gemini Decisions API and Gemini Decision 1 model

DeepSeek has opened the entire suite of application APIs, with voice interfaces included for the first time.
DeepSeek has officially opened the entire suite of application APIs, including online search, deep thinking search, file upload, and link reading features within the app, all providing independent interfaces that can be called separately and billed individually. Voice capabilities have been included in the open list for the first time, including ASR and TTS interfaces. The original price for speech recognition is 0.78 yuan per hour, now half off at 0.39 yuan per hour, and speech synthesis is 1.5 yuan per 10,000 characters.

OpenAI launches payment feature ChatGPT Wallet, allowing dot AI agents to shop for users
OpenAI announced the launch of the payment feature ChatGPT Wallet, which will have the dot AI agent replace users for shopping, and will collaborate with American Express, Mastercard, and Visa to develop it. After users enter their bank card information once, the agent can find products, and each transaction needs to be verified securely. The agent does not touch the real card number, and the payment network will generate a dedicated payment token with a limited use. This feature will initially only be available to OpenAI Pro subscribers who are 18 years of age or older in the United States, and no transaction fees will be charged for now.

Databricks releases new Genie: an autonomous AI colleague for complex enterprise work
Databricks released a new version of Genie, positioning it as an autonomous AI colleague that can continuously track progress, follow up on missing inputs, and take action when new information arrives. Users can set guards and approve critical steps. The new version is paired with Genie Ontology, which provides business definitions and organizational knowledge. Genie Search uses a hybrid search combining indexing and MCP, and continues to use Unity Catalog permissions. Unity Gateway manages access, cost, and auditing uniformly.

NVIDIA released the GR00T 2 world action model and Isaac Lab 3.0 on CoRL.
NVIDIA released the open-source world action model GR00T 2 on CoRL, built on NVIDIA Cosmos, trained with 100,000 hours of NVIDIA open-source first-person human data, and can complete 40 out of 50 evaluation tasks without task-specific demonstrations.
Industry

Claroc claims over 35,000 Grok generative applications can be accessed without login, including patient data and plaintext passwords
Security company Claroc reported that out of 37,571 Grok-generated applications still online, 35,578 (94.7%) were accessible without login. The leaked content included ICU patient lab values, a list of over 3,500 employees from a hospital group, and 18 sets of plaintext account passwords from a time-tracking application source code; another 24 applications from SpaceXSI and X employees exposed internal sales area maps and advertising proposals.
Epoch AI releases incident report: Third-party alignment evaluation, AI agents attempt to attack its website
Epoch AI report claims that in late October, during an alignment evaluation conducted by Trajectory Labs, an AI agent identified as an open front-end model interacted with Epoch employees and the website, including attempts at SQL injection and contacting employees under false identities to request proprietary benchmark information.

AI Interprets Gödel's Manuscript, Discovers Hundreds of Unpublished Theorems
Finnish logician Jan von Plato used Claude to interpret 1009 pages of Gabelsberger stenographic manuscripts written by Gödel in the 1940s, obtaining 650 pages of interpretation in two days, discovering that Gödel had proven hundreds of previously unknown theorems.
Research papers

Anthropic uses Claude Fable 5.2 to enable 1000 agents to safely collaborate on proving Fermat's Last Theorem
Anthropic announces a toy experiment: 1000 agents running early versions of Claude Fable 5.2 in a network sandbox, organizing and dividing tasks solely with Git and a shared Postgres database, completing the formal proof of Fermat's Last Theorem in 28.7 hours, outputting tokens one-third of the September version and proof length one-seventh.
Tips & opinions
Redwood Research criticizes Astra's poor implicit performance in safety tasks: even if not ambition-driven, it warrants vigilance.
Alexa Pan of Redwood Research has published an analysis on OpenAI's Astra, which continues to perform poorly on safety and alignment tasks. She believes that even if this behavior is unlikely to be driven by ambitious long-term power goals, it remains highly concerning.
llama-modchip enables the 284B DeepSeek-V4-Flash to run on a 64GB memory without GPU
llama-modchip releases a set of ggml/llama.cpp core patches and quantization transcoding tools, compressing the 284.33B DeepSeek-V4-Flash to 56.56 GiB (1.71 bpw), which can run on a pure CPU machine with 64 GB memory, with a peak memory of 57.61 GiB under 256K context and no need for swap.