GPT-6 Astra hits OpenResearch, tops Terminal-Bench-Science over Fable 5.1
burny_tech · x · 2026-09-06
alphaxiv reports that GPT-6 Astra is now available on OpenResearch: update Codex CLI to 0.153.1+, select the model, and let the autoresearch agent run.
- GPT-6 Astra is state-of-the-art on Terminal-Bench-Science 0.1, surpassing the recently released Fable 5.1
- The author notes Terminal-Bench Science was among the first benchmarks OpenAI and Anthropic reported in their model release blogs, making it the new litmus test for model quality
- Takeaway: agentic scientific research is becoming more real with every model release
More from Models
- Benchwarmer tool re-renders AI benchmark charts, recomputes winners from raw numbers — aronchick · 2026-09-06
- Programmer hasn't written code in months: Eleanor Berger on GPT-6 Astra and 'factory mode' — intellectronica · 2026-09-06
- Dev prompts Astra from his car, nails it in one shot after waiting two years — generativist · 2026-09-06
- GPT-6 Astra called best vision model yet, dethroning Gemini on detection — jonstephens85 · 2026-09-06
- Dev: Codex personal OAuth finally fixes 1M-token context — menhguin · 2026-09-06
- Leak: Grok Bot onboards ~100 customers as xAI sales enablement kicks off next week — Sauers_ · 2026-09-06