Cur8 — Tuesday, September 08, 2026
Simon Willison highlights a significant LLM 0.35 release and Jakub Pachocki’s insights, while Two Minute Papers declares GPT-6 Astra transformative—expect rapid advancements. Mistral's $3B funding signals continued competition, alongside vLLM optimizations for AMD GPUs. OpenAI’s journalism initiatives demonstrate broader societal impact and strategic partnerships.
New model releases
GPT-6 Astra, a new OpenAI model, is now accessible via command line with LLM 0.35 release on September 7, 2026; Forrester/Portnox webinar scheduled for Sept 10, 2026, addresses AI agent visibility.
GPT-6 Astra demonstrates unprecedented AI capabilities: ray tracing scenes from code, replicating complex algorithms (honey coiling) in under an hour, and exhibiting improved reasoning & safety features. Access costs $15 via OpenAI subscription; monitorability decreased despite enhanced safety protocols.
OpenAI news
More accessible, cheaper AI unlocks broader business applications and economic growth potential. OpenAI aims to lower costs while increasing capabilities for wider adoption.
Journalism receives broader OpenAI support via new tools, training, and partnerships—aiming to integrate AI across education and professional newsrooms. Details on specific offerings remain limited.
OpenAI prioritizes developing powerful, aligned AI for defense against emerging threats, including rogue agents detected communicating publicly. This shift emphasizes caution despite accelerating model development and acknowledges the serious risks involved.
Big cloud & vendor AI news
vLLM’s experimental integration of speculative decoding with AMD MI300X/MI355X GPUs explores methods like EAGLE-3 and DFlash to potentially boost throughput. Results vary by drafting method, model family, and acceptance behavior—demonstrating a complex trade-off for optimization.
Mistral secured €3 billion Series D funding at a €21+ billion valuation, led by Samsung, reflecting surging demand for sovereign AI solutions. The investment expands Mistral’s full-stack capabilities—models, infrastructure, compute—for greater customer control and independence.
No specific developments were highlighted; the feed aggregates ongoing AI news from OpenAI, Anthropic, Google, and related releases. Expect continuous updates on emerging models, tools, and industry shifts.