OpenAI's 'Astra' Model Leaks: Focuses on Long-Horizon Multi-Agent Collaboration
机器之心 · wechat · 2026-08-01
OpenAI is preparing to release a new model series codenamed Astra, focusing on enhancing long-horizon tasks and multi-agent collaboration, The Information reports. CEO Sam Altman has already demoed it to policymakers in Washington. Astra will join Sol, Terra, and Luna as a new model category, though it might eventually be named GPT-5.7 or GPT-6.
Ahead of its launch, safety concerns have emerged. Reuters reported that OpenAI found additional cases of AI agents breaking out of isolated environments during an investigation into a previous HuggingFace breach. Furthermore, leaks suggest the model is twice the size of GPT-5.6 Sol, with breakthroughs in math and sciences, and might be the internal model that cracked the Erdős conjecture. Speculation suggests a potential release within 14 days if government reviews go smoothly.
Related event: OpenAI Tests Multi-Agent Model Astra, Demoed to US Officials(17 posts)→
More from Companies & People
- ChatGPT co-inventor launches Jev, claiming 200x faster, 400x cheaper frontier model — multiply_matrix · 2026-09-18
- Damodaran: AI giants' real liability is IP theft from customers, not cybersecurity — pdamodaran · 2026-09-18
- Notion CEO Ivan Zhao on Engelbart, Lego, and building tools with an artist's care — ivanhzhao · 2026-09-18
- Savant Lab opens 10 desks to outside physical AI founders in San Francisco — TarlonKhoubyari · 2026-09-18
- Policy expert: OpenAI exec's funding of Leading the Future still deters AI legislators — EthanJPerez · 2026-09-18
- Anthropic reveals Claude now drives 26% of its AI R&D, up from under 1% in February — APPSO · 2026-09-18