Google DeepMind introduced new Gemini models for fast agentic workflows and complex knowledge work, while AWS made Claude Fable 5.2 and Mythos 5.2 available on its platforms. Anthropic expanded Claude into Google Meet, and AWS and NVIDIA announced infrastructure and tools for agents and robotics. Industry news included Anthropic’s planned IPO and a report that Grok Build apps exposed sensitive data through public links.
Google DeepMind Introduces Gemini 4 Hydrogen for Fast Agentic Workflows
Google DeepMind introduced Gemini 4 Hydrogen, a model for rapid execution across everyday coding and agentic workflows, available starting today across its developer, enterprise, and consumer products. Google says it scored 51 on the Artificial Analysis Intelligence Index at its default high thinking level and costs $1.50 per million input tokens and $7.50 per million output tokens. In Google Antigravity, the new Orchestrator can switch between models; on Terminal-Bench 4.0, it used 24% less quota and completed tasks 19% faster on average than the comparison cited.
Model releases and updates
Section guideThe releases span DNA variant prediction, image generation and editing, robotics, and evaluations of models on multimodal agent tasks.

Google DeepMind Makes Gemini 4 Carbon Available to Developers and Users
Google DeepMind has made Gemini 4 Carbon available to developers, enterprises, scientists, and consumers, positioning it for complex knowledge work, agentic coding, cybersecurity, and scientific reasoning. Paid developers and enterprises can access it through the API in Google AI Studio and Gemini for business at the same introductory price as Gemini 4 Argon: $2 per million input tokens and $10 per million output tokens. Google AI Ultra subscribers are also getting access through Google Antigravity, the Gemini app, and AI Mode in Search.

AWS Announces Claude Fable 5.2 and Mythos 5.2 Availability
AWS announces that Claude Fable 5.2 and Claude Mythos 5.2 are available on Amazon Bedrock and Claude Platform on AWS. Fable 5.2 is generally available for technical work, while Mythos 5.2 is limited to Anthropic verification programs for cybersecurity and life sciences. The post also covers Bedrock retention requirements, access options, and migration guidance using the claude-api skill in Claude Code.
Google DeepMind Announces AlphaGenome 2 for Cell-Specific DNA Variant Prediction
Google DeepMind announced AlphaGenome 2, a DNA sequence model that predicts how genetic variants affect specific human cell types and states. Trained on more than 107 million individual cell profiles, it reported nearly twice as many causal variant identifications as the original AlphaGenome on the TenK10K dataset. It is available for non-commercial research through the AlphaGenome API, with commercial research access in private preview.

ByteDance Releases Seedream 5.5 for More Stable Image Editing
ByteDance’s Seed team released the Seedream 5.5 image model on November 9, with editing that can keep unchanged areas pixel-level stable and generate images with transparent backgrounds. The post reports a severe text error rate of 0.4%, down from 2.8% in the previous generation, and says editing realism trails Nano Banana 2.1 by about 6%. Seedream 5.5 is rolling out across Jimeng, Doubao, and Doubao Work; developers can try Doubao-Seedream-5.5 on Volcano Engine Ark.
NVIDIA Introduces GR00T 2, a Robotics Model With Cross-Embodiment Skill Transfer
NVIDIA introduced GR00T 2 at CoRL in Austin, describing it as a world action model that predicts future video and robot actions together. The report details three paths for transferring demonstrations across humans, handheld devices, and robots, and cites success rates of 65% on Dexter and 60% on YAM for specific tasks. NVIDIA says 100,000 hours of human data are already open, with additional model, data, simulation, and collection resources to follow.
MiniMax Releases Open-Source H3-Image With Native 4K Image Generation
QbitAI reports that MiniMax has released H3-Image, an open-source image generation model that supports native 4K output at 4096×4096 pixels and up to 16 reference images. The author describes tests involving CAD views, exploded diagrams, academic text, and design assets, and says 4K images took about 10–20 seconds in the web interface. The article also reports API pricing at 50% to 70% of major competitors, with a further 50% discount on 16MP requests during the first week.
QbitAI Tests Tencent Hunyuan Hy4 on Multimodal Agent Tasks
QbitAI author Wen Le tests Tencent Hunyuan Hy4 through WorkBuddy on tasks including recreating a music player from a screenshot, building a 3D amusement park, creating and revising a 16-slide presentation, and making an image batch-processing tool. The article describes Hy4 checking its work, fixing issues, and rerunning tests during these tasks. It also reports internal evaluation results and says Hy4 is available through WorkBuddy and CodeBuddy for a limited free trial through November 15 at 23:59.

AINews: Claude Fable 5.2 Leads Benchmarks as Anthropic’s IPO Roadshow Begins
Latent Space’s AINews issue reviews Anthropic’s release of Claude Fable 5.2 and Mythos 5.2 alongside the start of Anthropic’s IPO roadshow. It reports Fable 5.2’s 16.4% score on SemiAnalysis GDPx, while the cost per task rose to $88.15 from $67.16 despite lower per-token prices. The issue attributes the benchmark result in part to more extensive test generation and subagent checks, and surveys other evaluations, access limits, and reactions.
Product releases and updates
Google Expands Antigravity Teamwork for Research and Engineering
Google says researchers and engineering teams used Gemini 4 models with Antigravity’s /teamwork to tackle open problems and production tasks, reporting results across formal proofs, molecular structure prediction, software rewrites, and TPU kernel optimization. Researchers and developers can use Gemini 4 with /teamwork and /boost in Antigravity today; Google also opened applications for six months of free access to its highest usage tier for 10,000 academic researchers.
Anthropic Introduces Claude Meetings for Google Meet
Anthropic is rolling out Claude Meetings in beta for Team and Enterprise plans, adding Claude as a participant in Google Meet calls. It can produce meeting notes, answer questions, and create documents and whiteboards during the call; notes, transcripts, and ordinary questions are included during the beta, while connected tool use counts toward the user's allowance. Anthropic says other meeting platforms are coming soon.

NVIDIA Contributes OpenShell and Adds AI Factory Infrastructure
NVIDIA is contributing OpenShell, an open source runtime for securing and governing autonomous agents, to the CNCF Sandbox and contributing to the Google-led Agent Substrate project. It also announced NVIDIA Kubernetes Engine and NVIDIA Kubernetes Recipes, both in public preview, and NVIDIA Snapshot. In evaluations reported by NVIDIA, Snapshot reduced GLM-5.3 workload readiness from roughly eight minutes to 31 seconds and vLLM startup times from two to 8.5 minutes to six and 34 seconds.

NVIDIA Introduces GR00T 2 and Robotics Simulation Tools at CoRL
NVIDIA introduced GR00T 2, an open world action model, alongside robotics simulation tools, benchmarks and agentic workflows at CoRL. The post reports that GR00T 2 completed 40 of 50 evaluated tasks at least once without task-specific demonstrations and achieved 95% average success with five demonstrations per task. Isaac Lab 3.0 with Newton achieved up to 2.5x the training throughput of Isaac Lab 2.3 with PhysX in a vision-based dexterous manipulation benchmark.

AWS Announces Faster, Larger Amazon Bedrock AgentCore Runtime
AWS announced updates to the Amazon Bedrock AgentCore runtime, including sessions that start in under 2 seconds, scale up to 8 vCPUs and 32 GB, and support container images up to 10 GB. The runtime can reclaim unused memory and let idle agents scale to zero, while sessions retain hardware isolation. The updated runtime is available today in AWS Regions across North America, South America, Europe, and Asia Pacific.
Industry
Jensen Huang Resigns as Nvidia CEO Amid Scandal and Investigations
Nvidia confirmed in an SEC filing that its board accepted Jensen Huang’s resignation, and CFO Colette Kress will serve as interim CEO while continuing her CFO duties. The timeline also recounts reports of Huang’s relationships with Megaspeed executive Alice Huang and Nvidia executive Kimberly Powell, as well as investigations into Megaspeed over suspected chip diversion to China. Nvidia, Huang, and Powell have not confirmed the report about Powell.
Grok Build Apps Exposed Sensitive Data Through Public Links, Report Says
QbitAI reports that security firm Claroc found more than 35,000 Grok Build apps accessible without login, with an estimated 650–800 containing private data. Examples included SpaceXSI and X sales information, Tesla employee performance and bonus details, ICU patient records, and plaintext passwords. Claroc said passive DNS databases could reveal app URLs; Ars Technica reported that 24 apps created by SpaceXSI and X employees had been taken down by publication time.

Microsoft Researchers Show How Dormant Neurons Can Backdoor AI Agents
Microsoft researchers describe Lazarus, a technique that reprograms fewer than 10 dormant neurons in an LLM to make an agent execute a malicious command when a hidden trigger appears, without retraining. The model can behave normally and pass standard evaluations until triggered. Microsoft Foundry uses model sourcing controls and scans surrounding software, while the researchers are developing weight-scanning and red-teaming defenses; they say reliable detection depends on comparing weights with the original.
Anthropic Sets IPO Price Range at $70 to $78 per Share
Anthropic plans to offer 825 million Class A shares at $70 to $78 each, implying a valuation of up to $1.404 trillion and estimated IPO proceeds of $57.75 billion to $64.35 billion. Its updated filing also reports $81 billion in run-rate revenue by the end of October, a nine-month net loss of $104 billion, and $900.9 billion in compute and infrastructure commitments.
Research
Google DeepMind Uses AlphaProof Code to Verify Signal Protocol Code
Google DeepMind researchers used AlphaProof Code to formally verify key security properties of the compiled code underlying the Signal protocol. The multi-LLM framework generated more than six million lines of Lean proofs, which Lean checked automatically; human experts reviewed every specification and focused on one percent of the codebase flagged for a second look. The process took four weeks, and tests with two previously fixed bugs produced machine-checked proofs that the buggy versions were insecure.
Tips and perspectives
Stanford CS329Z Publishes Agent Engineering Lectures and Two Assignments
Stanford’s CS329Z course is making its Engineering AI Agents materials public, with lectures appearing daily on Stanford Online’s YouTube channel and two MIT-licensed assignments on GitHub. The first assignment has students build an agent harness from scratch and includes a test suite; the second teaches students to create and evaluate a Terminal-Bench-style task. The post also describes the assignments’ API costs and requirements.
- 2026-W45Google DeepMind Rolls Out Gemini Omni 1.2 Flash for Longer Video Generation
- 2026-W44Mistral releases improved Mistral Large 4 open weights for coding and cybersecurity
- 2026-W43Mistral introduces Voxtral 3 speech models and a real-time voice agent SDK
- 2026-W42Meta releases Muse Spark Open, an open-weight multimodal mixture-of-experts model
- 2026-W41OpenAI Releases Nearly 400 Mathematical Results at Once, Mathematicians Say It Will Take Years to Digest
- 2026-W40OpenAI Pauses Training as Gemini 4 Argon Takes the Lead
- 2026-W39e/acc Weekly · 2026-W39
