Good morning, welcome to this week’s Lore Brief, your 3-minute brief of the most important moves in AI and tech.

This issue is brought to you by Factory, the fastest way to ship software with autonomous engineering agents.

  • ChatGPT answers can now be apps → GPT-6 is rolling into the Chat tab with Intelligent UI, so a reply can include charts, buttons, forms, and small tools such as a budget calculator. Paid plans get it today on GPT-6 Sol. Free and Go start tomorrow on Luna. Work and Codex are not part of this switch. Read more here →

  • Grok Bot will call other companies’ models → Elon Musk said SpaceXAI will route Grok Bot jobs to whichever model is most likely to finish them best, including Claude Opus 5.5, Midjourney, and Suno. The Bot product is the one that gets real work, not the public Grok LLM chat. Read more here →

  • Isomorphic is talking about a $40 billion round → Bloomberg says Demis Hassabis’s drug-discovery lab, spun out of DeepMind, is in early talks to raise at a valuation of at least $40 billion, and possibly $50 billion. The round is not closed. Five months ago the company raised $2.1 billion, with Alphabet supplying a large share. The business exists to turn AlphaFold-style protein work into drugs. Read more here →

A ring to control your AI agents:

State of AI:

  • Claude Haiku 5.5 is about 75% cheaper to run → Anthropic’s small model is a clear step up on coding, computer use, and office work, and it is the first Haiku with a dial for cost versus intelligence. Prompts under 100,000 tokens, about 90% of old Haiku traffic, are priced about 90% lower. It is live on Claude, AWS, Google Cloud, and Azure. Cache reads on Sonnet 5.5 were also cut in half, to $0.10 per million tokens. Read more here →

  • OpenAI publishes 722 math papers, 162 machine-checked → An unreleased internal model produced 722 manuscripts from about 4,000 attempts, at roughly three hours of ChatGPT Pro compute each. 162 have a Lean proof. The other 560 are not machine-verified, and OpenAI says those can still be wrong. Loud posts claimed a Hodge proof and Riemann progress. Read more here →

  • DHH’s code review is just “ask the other model” → The Rails creator said he tells Codex to review a change with Claude, and Claude to review one with Codex. The models already know how to start the other tool, take turns, and settle a disagreement. No new product. The post is a working habit that picked up a few thousand likes from people doing the same thing. Read more here →

  • Paid ChatGPT can transcribe a recording → OpenAI’s October 6 notes say paid plans can upload an audio file and ask for a transcript, a summary, or answers about what was said. The cap is 512 MB, in formats including MP3, WAV, and M4A. Free does not get it. Some users said they had already done this on Plus, so the new part is file upload in chat, not voice itself. Read more here →

  • xAI published a Bot instruction manual → A new guides page tells people how to set up Grok Bots for real jobs. Ben Lang’s tip, which passed half a million views, is to point your main Bot at the page and have it recommend a setup. Early replies said the first suggestion was simply to run fewer Bots. Read more here →

That’s it for this week’s Lore Brief.

See you next week!