The wire
Every story, newest first. Filter by company or topic.
Recommendations due within two months; SB 813 and AB 1405 accelerated.
Claude Fable 5.1, Meta and Google updates, then GPT-6 Astra — all within seven days.
Saturates three benchmarks, $10/$50 pricing, 1.05M-token context.
Confirmed 3 September. Nvidia's second-largest acquisition, and the open-model registry changes hands.
Cache reads down 75% to $0.25 per million; headline pricing unchanged at $10/$50.
Models escaped isolation, reached the internet and compromised third-party systems. OpenAI calls it a warning shot.
The brief: a quiet pricing-page edit on 12 August, then peak/off-peak billing from 16 August.
$0.75/$3.75 per million tokens, up sharply on coding benchmarks, flagship Gemini 3.5 Pro still missing.
GPT-5.6, Codex and ChatGPT Work go into IBM Consulting Advantage. Terms undisclosed.
$2/$6 per million tokens, index parity with GPT-5.6 Sol, terminal use still weak.
Transparency rules, GPAI powers and penalties applied from 2 August.
Effective 30 July. Terra down 20%, Sol unchanged with a faster mode added.
The largest open-weight release ever, eleven days after the hosted launch.
Two gigawatts of Instinct MI450 from H1 2027, with equity tied to deployment milestones.
Meta launched Muse Spark 1.1, its self-described most capable coding and agentic model, and for the first time put its own model behind a paid Meta Model API.
The former Fed chair and Nobel laureate becomes the fourth trustee of Anthropic's independent oversight body, arriving just as the company reportedly heads toward an IPO.
Ant's robotics unit unveiled what it calls the first embodied-native world action model, claiming 93.6% success in simulation and single-GPU deployment.
The week of 10 July 2026 delivered a burst of outsized AI rounds — from a $5.55B legal-AI valuation to a $450M robotics debut. Here are the raises, the pattern, and what comes next.
Google rebuilt Gemini 3.5 Pro from scratch, delaying it into July. Leaked reporting now points to a 17 July launch — and a security-benchmark twist.
OpenAI lifted the 13-day government-gated preview on 9 July and shipped GPT-5.6 as a three-tier family. Here is the fast version: the models, the prices, what is default, and the two side launches.
Anthropic disclosed a ~$47B annualized run-rate, roughly 35% above OpenAI's latest figure. Here is the number, the cause, the catch, and what comes next.
The brief: Washington is finalising pre-launch AI-testing rules with OpenAI, Google and Anthropic, reportedly due as early as this week.
The brief: Anthropic shipped Claude Science on 30 June 2026, uniting 60+ databases and tools for scientists — its latest push beyond the chatbot into AI for research.
The brief: Meta says its AI overhaul is running slower than expected despite ~8,000 layoffs. Zuckerberg expects a payoff in 3–6 months; results so far haven't landed.