Anthropic says Haiku 5.5 targets high-volume work and is its first Haiku with an effort setting. Simon Willison says it costs $0.10/$0.50 per million tokens up to 100,000 tokens, then 5x more.
Why it matters: Willison says it uses about 1.25x as many tokens as Haiku 4.5, and that above 100,000 tokens GPT-6 Luna looks like a much better deal.
The Verge says ChatGPT answers can now mix text with charts, forms and buttons. Plus, Pro, Business and Enterprise users get GPT-6 Sol now; free and Go users get GPT-6 Luna on Thursday.
Why it matters: Answers that build tools in-line change what users expect from chat products. TechCrunch says users who want fewer visuals can dial them back.
Nvidia says Microsoft's MXC lets agents run in the background under OS control. The Verge says Copilot will soon find, zip and email local files, and the Surface Laptop Ultra starts at $2,599 on October 16.
Why it matters: Windows is turning into a runtime for agents that act on your files, so who approves those actions becomes an IT question, not just a product one.
Per a summary of the report, Microsoft cut most monthly AI spending caps from $100,000 per employee to about $10,000, and Meta's Claude Code users fell from around 60,000 to about 30,000 as it moves to in-house tools.
Why it matters: Even the largest buyers are capping coding-agent spend, a sign AI tool budgets now get the same scrutiny as any other line item.
CNBC says Samsung's preliminary Q3 operating profit of 107.4 trillion won is 782% higher than a year earlier, on revenue of about 195 trillion won. Its shares still fell 0.7%.
Why it matters: Memory is a bottleneck for AI; an eToro analyst told CNBC buyers are signing multiyear supply deals as agents consume more memory.
TechCrunch says Robot Ventures led the Series B, with Nvidia and others joining. Nous estimates its open-source Hermes Agent drives about 2.5% of global AI token usage; the new product deploys custom agents for companies.
Why it matters: Open-source agent makers are now selling to enterprises directly, giving buyers another option beside the closed labs' platforms.
TechCrunch says testers saw just two break reminders across nearly 2,000 prompts, and the bot pointed to a trusted adult in 94% of crisis prompts about risks from others. OpenAI says testing may predate parental controls.
Why it matters: Engagement design in teen chatbots is drawing lawmakers' attention, including the bipartisan CHATBOT Act that TechCrunch cites.
Wired says the DOD's Tradewinds program gives accepted pitches 'post-competitive' status, sometimes enabling awards in under a week. This year it seeks agents for joint targeting; OpenAI, Anthropic and Google are listed.
Why it matters: Faster defense buying pulls AI labs deeper into military work, while Sen. Joni Ernst tells Wired the contracts used raise oversight red flags.
TechCrunch says the platform reads records nightly in more than 50 hospitals to screen for conditions like malnutrition and delirium. The company says it doesn't diagnose; it only highlights patients for review.
Why it matters: It shows a clinical AI pattern that keeps clinicians in charge, sold with outcome-based pricing tied to ROI the hospital validates.
Three engineers linked the model to cameras and power steering, with a safety driver ready to brake. On their DrivingBench course only Astra finished; Claude Fable 5.1 made it 45% of the way and Grok 11%.
Why it matters: General models are starting to act in the physical world, and the engineers told Wired careful prompting got past the models' initial refusals.
A source told CNBC the debt deal would lean on the investment-grade market and likely use GPUs as collateral. Talks are preliminary, and the Financial Times first reported them.
Why it matters: GPU-backed debt at this scale tests an assumption a credit source described to CNBC: that GPUs hold their value for about seven years.
The research firm values Anthropic at $150 billion against a reported $2 trillion target. It cites Reuters figures from a leaked prospectus: $4.6 billion in 2025 revenue and a $42 billion net loss.
Why it matters: The IPO will put a public price on a frontier lab. CNBC notes this bear was right on WeWork and wrong on DoorDash.
Our job-board count shows Databricks with 894 open roles, OpenAI with 815 (8 fewer than last run) and Anthropic with 647, where Sales is the largest team at 120 openings.
Why it matters: Go-to-market roles top OpenAI's list too, at 143, a sign labs are hiring hard to sell what they've built.
Our registry lists two EU AI Act dates on December 2: new bans on AI that makes non-consensual sexual deepfakes or child abuse material, and Art. 50(2) output marking for generative AI on the market before 2 Aug 2026.
Why it matters: Teams offering generative AI in the EU have under two months before these dates. This describes the registry's timeline; it is not legal advice.
Our laws registry lists November 28 as the deadline, 60 days after signing, for proposed legislative language defining 'Super Intelligence', per the order's Federal Register text.
Why it matters: A federal definition could shape which systems future rules cover. This describes the order's timeline, not what the language will say.
The Apache-2.0 repo, created October 2, offers open-source AI code review for Claude Code and Codex before you push: scanners on changed lines, then a separate reviewer process. It has 282 stars.
Why it matters: Checking agent-written code before it is pushed is a growing need as coding agents produce more changes.
The MIT-licensed agent skill answers hard questions with a one-page HTML file you can read. Created October 2, it now has 2101 stars.
Why it matters: Output format is becoming a product choice for agents, the same bet ChatGPT is making with visual answers.
The 'Sharing AI progress in mathematics' thread has 1241 points. On Bluesky, @aeporreca.org shared a statement from the Association for Human Mathematics, and @gracekind.net wished the results were easier for humans to follow.
Why it matters: How labs package machine-made proofs is becoming a credibility question with the experts who have to check them.
Bloomberg's Tech In Depth newsletter, by UK policy reporter Jacob Reid, looks at the study. Its Hacker News thread has 93 points and 32 comments.
Why it matters: Businesses routing shoppers through AI assistants will want to know whether advice shifts with a user's perceived wealth.
@SenSanders posted Quinnipiac figures, including 91% who say AI guardrails are important and 72% opposing new data centers in their community. @MTSlive posted that 69% of Americans want stricter AI regulation.
Why it matters: Public mood on AI rules and data centers is turning political, which can shape siting fights and policy that builders will feel.