Morning Edition · Friday, November 6, 2026
Tech
OpenAI rates its new GPT-6 Cyber model "Critical" for cyber risk and limits it to vetted security teams
OpenAI reports a 48.5% score on its ExploitGym benchmark against 32.1% for GPT-5.6 Cyber, along with working exploits for 948 Linux kernel bugs, and no outside lab has reproduced any of these figures.
Tech
Moonshot AI publishes the weights for Kimi K3.1 Pro, a 2.8-trillion-parameter mixture-of-experts model
The model activates 104 billion parameters per token and ships in a 4-bit format from quantization-aware training, and an independent index scores it 55, three points behind Claude Opus 5.5.
Markets
Anthropic moves toward a November roadshow for its Nasdaq listing with no price range filed yet
Its prospectus reports $16.3 billion in first-half revenue, 80% of it consumption-based, and an $80.7 billion net loss that includes a $72.4 billion non-cash charge.
Tech
OpenAI pledges to pay grid costs for its own data-center campuses as NVIDIA sells Vera CPUs on agents per watt
Both companies published their plans on November 5, two days after the US midterm elections, and both treat electricity, rather than chips, as the limit on how far agent deployments can grow.
Tech
OpenAI reports an unreleased model scoring 0.875 on BenchCAD 2.0, against 0.594 for GPT-6 Astra
The benchmark's maintainers have not listed the result, OpenAI has not published its evaluation settings or named the model, and the best public score remains Astra's 0.59.
Tech
Octopus paper shares one KV cache across sub-agents and lifts BrowseComp-Plus accuracy to 49.6% on a 4-billion-parameter model
The same week, a second paper made reproducible inference cheaper under batching, and Anthropic cut the cost of Claude Code's background subagent status updates by about 99% after the tool had re-read whole transcripts every 30 seconds.
Tech
Google cuts Gemini Omni video pricing 30% to $0.07 per second and extends clips to 15 seconds
Gemini Omni 1.2 Flash is now generally available in the Gemini API, Google Cloud and Google's own ad and productivity products, with provisioned throughput for production buyers.
Tech
Claude Code sent users' account email addresses to websites, and Anthropic answered with an opt-out privacy module
The fix arrived the same week Anthropic began a US beta that gives Claude read-only access to up to two years of users' bank transactions through Plaid.
Tech
Anthropic says Claude helped solve the first 3D structure of a vault particle from a bacterium
Claude searched about 15 billion predicted proteins, seven of 15 lab-tested candidates formed vault-like particles, and cryo-electron microscopy confirmed four, according to a preprint that has not been peer reviewed.
Tech
Figure posts video of new F.03 humanoids walking out of its BotQ plant without saying how they were controlled
The roughly 19-second clip shows robots crossing a pedestrian path with people nearby, but Figure gave no production total, destination or autonomy details.
Tech
New FAH hand-estimation models run at 65 frames per second on a consumer GPU and estimate contact without object geometry
The Apache 2.0 models fine-tune and distill the hand branch of SAM 3D Body, predict 2D and 3D points directly, and fit MANO or MHR-Hand meshes in a separate step.
Tech
OpenAI and NVIDIA publish agent case studies that name customers but report few measured outcomes
Merge says it uses GPT-6 Astra and Codex to review product code, Cult.fit says Codex lets teams outside engineering build tools, and an NVIDIA engineer used Astra with Omniverse libraries to build and debug a warehouse simulation.