The Polylog AI Intelligence Brief

Morning Edition · Wednesday, November 11, 2026

Tech

Google Opens Gemini 4 Carbon at $2 per Million Input Tokens, a Quarter of Anthropic's Fable 5.2 Price

Cursor reports Carbon scoring 60.4% on CursorBench against 60.0% for Fable 5.2 Max, but the introductory rate doubles on December 1, and Google's claim of lower cost per task is not backed by published data.

6 sources
Plausible

Tech

Anthropic Ships Claude Fable 5.2 With a 20% Price Cut and Keeps Its Cyber-Capable Mythos 5.2 Behind Vetting

Fable 5.2 scores 65 on the Artificial Analysis index, the top score among 74 models, but costs $26.54 per index task, and its counterpart Mythos 5.2 found 89.6% of known vulnerabilities on a rediscovery test.

5 sources
Plausible

Markets

Anthropic Sets Share Price Range of $70 to $78, Valuing It at Up to $1.405 Trillion

The 825-million-share offering would raise between about $57.8 billion and $64.4 billion, and reported investor valuation talk has fallen from an earlier range of $1.5 trillion to $2 trillion to a more recent $1.1 trillion to $1.3 trillion.

4 sources

Tech

DeepMind's AlphaProof Code Agents Produce Lean Proofs That Signal's Compiled Encryption Code Meets Its Specification

Agent teams produced millions of lines of machine-checked proof in about four weeks against roughly 26,000 lines of human-written specification, and the checks correctly failed on two reintroduced bugs.

3 sources
Corroborated

Tech

Gemini Robotics 2.5 Reaches 96% Success Sorting Metal Parts in a Mercedes-Benz Pilot With Apptronik

DeepMind reports success rates 30% to 100% above its prior baselines across more than 200 tasks and released an adversarial safety benchmark, but the model goes only to robotics partners.

3 sources

Tech

Meta's Muse Ember Matches Claude Haiku 5.5 on Artificial Analysis at About One-Third the Cost per Task

Meta's smallest model costs about $0.14 per index task against $0.41 for Haiku, but trails Haiku by more than seven points on FrontierCode 2.0 and ranks thirteenth of 36 on the independent Vals Index.

4 sources

Tech

MiniMax Releases Open Weights for a 33-Billion-Parameter Model That Generates 16-Megapixel Images, With a License Reported to Exclude the US, EU and UK

H3-Image generates 4096-by-4096 images in six distilled steps without classifier-free guidance, and its license restrictions follow a pattern that MiniMax tied to Hollywood copyright litigation.

4 sources
Corroborated

Tech

Berkeley Lab Runs Meta's SAM 3 and DINOv3 on 300 A100 GPUs to Label X-ray Scans in About 15 Minutes

New arXiv research on auditable human-agent teams for X-ray experiments addresses the harder problem: sustained autonomous operation of physical instruments.

4 sources

Tech

NVIDIA Hands Its OpenShell Agent-Sandboxing Runtime to the Cloud Native Computing Foundation

Red Hat, Docker and Canonical co-maintain the runtime, which keeps credentials out of agents and enforces YAML access policies, and it enters the foundation as a Sandbox-stage project at KubeCon in Salt Lake City.

4 sources

Tech

ChatGPT Holds About 81% of Assistant Traffic as OpenAI Launches Merchant Listings Inside the Chatbot

Graphite's estimate gives Claude 6.8%, Gemini 5% and DeepSeek 4%, and OpenAI's new Businesses in ChatGPT tool lets companies manage how the assistant presents them.

3 sources
Plausible

Tech

Google Cloud Launches Gemini for Healthcare Aimed at Clinical Data and Administrative Workflows

The same-day launches target hospital administration, not clinical decisions, and Google added Gemini features on more Android devices and Adobe tools inside Drive and Gemini.

4 sources

Tech

Apple's Xcode Plugin Joins Claude Code's Official Directory as Claude Sheets Enters Paid Beta

The plugin lets Claude Code build, run and test iPhone apps through the Xcode binary installed on each developer's Mac, and Anthropic is adding spreadsheet generation to its paid plans.

3 sources
Plausible