#modelrelease

OpenAI dumped 722 AI-generated math proofs on GitHub. Nearly all came from one prompt to a single AI agent.

Google's new Gemini 4 Argon can reportedly hunt and patch software vulnerabilities on its own, and Google says it tops OpenAI's GPT-6 Astra and Anthropic's models on a third-party index. Lovely. That's the fourth 'best AI in the world' this month.

A new foundation model that doesn't generate text reveals what some builders think generative AI got wrong.

OpenAI gave ChatGPT Voice plugins, a choice of GPT-6 Astra, Sol or Luna, and a seat inside ChatGPT Work. Voice is now a routing and permissions decision you own

A 40% bigger base model at unchanged pricing. The headline benchmark is contested, and token burn may erase the saving. What builders should test this week.

Shanghai AI Lab dropped a 744B agentic model under MIT license with no blog, no paper, no PR. The strategy matters more than the model.

OpenAI shipped GPT-6 Sol and GPT-6 Luna on 22 September 2026 at half the old mid-tier price. Permanent, not promotional — and it breaks every routing threshold

Claude Opus 5.5 lists at $4/$20 per million tokens, 20% below Opus 5. Anthropic's 40% figure is an effective-cost claim you have to measure on your own workload

A 744B open-weight model lands with no paper and no pricing. Five working days, cheapest gate first, to turn that into a defensible adopt, shortlist or reject.

A full-duplex voice layer that delegates reasoning to a text model is two systems, two meters and one conversation. Who owns state, what the handoff sends, what

OpenAI shipped a full-duplex voice model in the API on 10 September 2026 at $0.05 a minute, billed per second. It hears and talks at once — and it is a front en

The InternLM team released Intern-S2-397B under Apache 2.0 on 11-13 September 2026, a multimodal mixture-of-experts model aimed at scientific reasoning and long agent tasks; the FP8 weights are a 406.3 GB download and the recommended setup is a node