AI News

AI news brief: Monday, November 9, 2026

33 stories, most important first.

OfficialModels · · Updated

Anthropic's Fable 5.2 reportedly launches at $8 input and $40 output per million tokens

Commentator Matthew Berman says Fable 5.2 is out and about 20% cheaper than Fable 5.1, citing benchmark results he presents himself.

Why it matters: If the stated prices hold, teams using Anthropic models could see lower per-token costs for a model the commentator says leads several coding and computer-use benchmarks.

6 sources
ReportedChips & Infrastructure · · Updated

Nvidia opens probe into CEO Jensen Huang over alleged Chinese spy ties

Nvidia has opened an independent investigation into Jensen Huang after U.S. intelligence agencies alleged that a woman he had an affair with is a Chinese spy.

Why it matters: The probe adds uncertainty at the world's most valuable AI chipmaker, which sits at the centre of U.S.-China export control tensions. Companies that depend on Nvidia supply or partnerships face headline risk while the allegations remain unresolved.

5 sources
OfficialResearch · · Updated

Anthropic: 1,000 Claude Fable 5.2 agents reproduce a formal Fermat's Last Theorem proof

Anthropic reports that a team of 1,000 agents running an early Claude Fable 5.2 reproduced a formal proof of Fermat's Last Theorem.

Why it matters: The run shows large agent teams being applied to a formal mathematical proof, which suggests coordinated agent fleets could handle long, verifiable technical work. Companies weighing such setups should note that Anthropic presents the shared database and layered safety controls as central to the approach.

3 sources
OfficialModels ·

Google DeepMind's Gemini Robotics 2.5 targets robot dexterity, reliability and safety

Google DeepMind describes Gemini Robotics 2.5 as its most capable vision-language-action model, aimed at improving robot dexterity, reliability and safety.

Why it matters: Teams building physical AI or robotics products have a new model from Google DeepMind to assess, though the claims so far are general and unquantified.

1 source
ReportedModels ·

NVIDIA's Nemotron 3.5 Bolt targets local agents within a 24GB memory budget

NVIDIA's Nemotron 3.5 Bolt is a 74B-total, 9B-active vision-language model designed to run on a 24GB memory budget with tool calling.

Why it matters: Teams building local agents can get visual understanding and tool calls without sending inference to a hosted API, but 24GB is the lower end of what the source describes.

1 source
ReportedChips & Infrastructure ·

Google takes over Crusoe's troubled Wyoming data center project

Google paid Crusoe an undisclosed sum to take over an unfinished Wyoming data center, alongside a $17B development commitment.

Why it matters: Hyperscalers are buying direct control of AI capacity to avoid delivery delays, which may tighten supply for companies that depend on third-party data center builders.

1 source
OfficialAI in Practice ·

Anthropic says Claude designed and prototyped three devices with Diode Computers

Anthropic reports that Claude Fable 5.2 designed circuit boards, enclosures and firmware for three devices, and that Diode built working prototypes without circuit fixes.

Why it matters: According to Anthropic, an AI model can now take a hardware design from concept to a working prototype with little guidance, which could shorten hardware development cycles. The results still come from a vendor-run test with early prototypes, so certification and production readiness remain open.

1 source
ReportedProducts ·

llm-anthropic 0.31 adds Fable 5.2 and Mythos 5.2; max effort returns no image

The llm-anthropic 0.31 plugin adds Anthropic's Fable 5.2 and Mythos 5.2, but a max reasoning effort test ran out of tokens without returning an image.

Why it matters: Teams that set maximum reasoning effort on Fable 5.2 may get empty outputs after heavy token use, which can break pipelines that expect a response.

1 source
ReportedProducts ·

OpenAI makes ChatGPT Enterprise available through AWS Marketplace

OpenAI says buyers can now purchase ChatGPT Enterprise through AWS Marketplace using their existing AWS purchasing and billing relationships.

Why it matters: Companies that already buy through AWS may be able to add ChatGPT Enterprise to their existing purchasing and billing setup instead of managing a separate vendor contract.

1 source
OfficialResearch ·

Core Automation trains one MoE model across two clusters 9,000 miles apart

Core Automation reports training a roughly 32B-total mixture-of-experts model across two clusters about 9,000 miles apart, using different accelerators.

Why it matters: If distributed training across mixed hardware matures, smaller groups may build large models without a single hyperscale data center. The company argues that power, grid equipment and land, rather than chips, are the main bottlenecks.

1 source
ReportedProducts ·

Meta adds Spaces, persistent project workspaces for its Muse agent

Meta introduced Spaces, project-specific work areas where its Muse agent can maintain planners and trackers.

Why it matters: Persistent agent workspaces move assistants from one-off chats toward ongoing project work, which raises questions about data handling and oversight.

1 source
ReportedBusiness & Funding ·

Investors look to India as a hedge against the global AI selloff

The Nifty 50 gained nearly 9% over the past month while AI-heavy markets in the U.S., Japan, South Korea and Taiwan fell.

Why it matters: Index composition explains why India offers limited direct AI exposure, which matters for firms weighing where to allocate capital around the AI cycle.

1 source
ReportedModels ·

NVIDIA's Nemotron 3.5 report details memory, decoding and compression gains for agents

NVIDIA's Nemotron 3.5 report says hybrid layers, draft-token decoding and compression speed up agent models, citing 2.43x throughput for the four-bit Super version.

Why it matters: Companies running multi-call coding agents may see lower serving cost per request, but the reported throughput gain does not directly translate into faster completion of agent tasks.

1 source
OfficialResearch ·

Richter et al. find aligning cell data types does not fully integrate them

A Nature Machine Intelligence paper by Richter et al. reports that aligning different biological data types does not fully integrate them in cell foundation models.

Why it matters: Teams building multimodal biology models may need to test whether each added data source contributes new information rather than duplicating existing signal.

1 source
ReportedModels ·

Artificial Analysis says Meta's Muse Ember is the cheapest model scoring 36

Artificial Analysis says Muse Ember scores 36 on its Intelligence Index, matching Claude Haiku 5.5 (max) at about a third of the cost per task.

Why it matters: For agent workloads where cost per task drives the budget, Muse Ember offers a similar score at a fraction of the price of comparable models. Its low answer rate, though, means teams that rely on factual recall should expect more declined answers.

1 source
OfficialProducts ·

Claude Code 2.1.321 changes gateway upstream URLs and layered MCP policy

The release makes Bedrock, Vertex and Foundry upstreams ignore ANTHROPIC base URL variables and adds up managed MCP servers under layered policy.

Why it matters: Operators of a Claude apps gateway may see changed routing and more managed MCP servers served after upgrading, which affects access control.

1 source
OfficialResearch ·

Artificial Analysis Endpoint Accuracy Index v1.1: 24 of 103 endpoints trail reference

Artificial Analysis says 24 of 103 serverless endpoints score statistically below its self-hosted reference in the updated Endpoint Accuracy Index v1.1.

Why it matters: Two endpoints serving the same open-weights model can differ noticeably in accuracy, so price and speed alone are not a sound basis for choosing a provider. According to the company, tool-calling and output-length settings are the most common sources of loss.

1 source