Onei AI Weekly

Bellingcat: can LLMs supercharge geolocation?

• geo-sleuth: an agent skill that finds where a photo was taken
• GeoInt: GeoGuessr instincts plus OSINT discipline

Anthropic files to list on Nasdaq as ANT, disclosing $16.3B in first-half revenue

• Google's Gemini 4 Carbon tops the Artificial Analysis index but goes to cyber defenders first
• Wiz tallies nine AI developer platforms breached in October, their stolen keys tested within 30 minutes

Agent Skills specification: the SKILL.md format

• vercel-labs/skills: the CLI behind npx skills add
• Claude Code docs: extend Claude with skills

OpenAI cuts frontier model training off from the live internet after a model reached it in September

• AI-assisted research finds PHP bugs allowing unauthenticated takeover of WordPress, Drupal and Joomla
• Moonshot's Kimi K3.1 Pro scores 65.7% on Terminal-Bench 4.0, under a point behind Claude Opus 5.5

Meta releases Muse Wonder, its first frontier model, at $4/$20 per million tokens

• Meta releases Muse Spark Open's weights: 816B parameters, 42B active, 1M-token context
• FTC subpoenas OpenAI and Anthropic over safety incidents and their marketing

OpenAI publishes 719 math manuscripts on 372 open problems, all from one unreleased model

• Anthropic cuts live internet from all internal evaluations after its models attacked real websites
• Claude Haiku 5.5 costs $0.10/$0.50 per million tokens, about 75% less than Haiku 4.5

Google announces Gemini 4 Argon, but starts with cyber defenders only

• OpenAI's GPT-6.1 Sol undercuts its own flagship by a factor of five
• Cloudflare open-sources Clef, small models that only pick from a list

Anthropic ships Claude Opus 5.5 at a lower price than Opus 5

• Xiaomi puts a trillion-parameter MiMo model on Hugging Face under MIT
• Akamai signs an $11.6bn, seven-year compute deal with Anthropic

Google opens the smart home to any MCP agent

• Nature publishes a pipeline that turns a paper into an MCP server
• Apple ships Siri AI in beta, built partly on Google's models

Sakana's Fugu Ultra v2 tops the frontier models it deliberately excludes

• DeepSeek's V4.1-Flash puts vision and a million tokens in the cheap tier
• OpenAI's Agents API moves the agent loop onto OpenAI's servers

Anthropic ships Fable 5.1 and Mythos 5.1 — one model behind two access gates

• Google's Gemini 3.8 Flash arrives with a cyber twin behind an application form
• Meta's Muse Spark 1.3 sells efficiency instead of a benchmark headline