·
Finding signal on Twitter is more difficult than it used to be. We curate the best tweets on topics like AI, startups, and product development every weekday so you can focus on what matters.
PLUS: Anthropic's cheaper Claude Haiku 5.5
OpenAI is releasing a new version of GPT-6 to all ChatGPT users, scaling it to 1.2 billion people.
Anthropic released Claude Haiku 5.5, its cheapest, fastest, and most capable small model, at around 75% less cost than Haiku 4.5.
ChatGPT now answers with fully interactive user interfaces, and GPT-6 is coming to everyone.
Greg Brockman introduced Intelligent UI, a ChatGPT capability that answers quickly with fully interactive interfaces.
A ChatGPT team member says the interface gets built around your prompt, from charts to diagrams to disposable tools.
Intelligent UI reaches paid ChatGPT users today and free users starting tomorrow, adding visuals and interactive elements to text replies.
OpenAI's launch videos show the interface building itself from a simple prompt, with a different shape each time.
Claude Haiku 5.5 is the first Haiku to support effort, priced at $0.10 and $0.50 up to 100k tokens.
An Anthropic employee says Claude Haiku 5.5 is now available and soooo good in a launch-day post.
OpenRouter measured Claude Haiku 5.5 at 100+ tokens per second and about 75% cheaper than its predecessor.
Claude subscriptions now include monthly API credits by plan, with Max 20x users getting $200 each month.
Anthropic's Claude Haiku 5.5 is live, according to a launch-day alert.
Theo rewrote the TypeScript compiler in Rust, saying ~$20k of Opus got there in 2 weeks after ~$400k of Codex tokens failed.
The US government, Google, Meta, and Biohub are launching a $1.8B push for AI-ready biological data.
Google opened a public portal where anyone can check files for a SynthID watermark, including content from OpenAI and Nvidia.
OpenAI says ChatGPT for Teens applies automatically to under-18 accounts, with protections on by default and a College Planner coming soon.
Mathematicians facing OpenAI's hundreds of AI proofs describe problems solved in inhuman ways that raise questions about what knowing means.
A developer built an interactive explorer for OpenAI's 722 math abstracts, with search, maps, and Lean coverage notes.
A DeepMind researcher says confidential compute enables double blind evaluations where labs run benchmarks they provably cannot see.
Andrew Trask expects millions of models routed per prompt to beat any single frontier model on quality and price.
NVIDIA's VERA paper shows a 27B agent that co-evolves its model and harness scoring 71.6 on AutoCoWorkBench, above Claude Opus 4.8.
Google's Gemini 3.8 Live ranks first on the Speech to Speech Index, with the standard version at $0.84 per hour of input audio.
The US grid plans to add 86 gigawatts this year, with zero coming from new nuclear, according to a16z.
Texas's grid maxes out only about 200 hours a year, so flexible data centers could add 100 GW without new plants.
Theo's agents wrote over 200 bad watch-PR scripts in a few months, and he shared a prompt to audit the waste.