GPT-6 Astra tops Vals Science Index at 65% accuracy while using 22x fewer tokens than Claude
haider1 · x · 2026-10-19
Vals AI released its Science Index measuring AI's ability to do scientific research. GPT-6 Astra leads with 65.09% accuracy at $8.03 per test; GPT-6.1 Sol follows at 59.63% for just $1.36, with Claude Sonnet 5.5 (59.12%, $13.24) and Opus 5.5 (58.65%, $9.13) behind. Haider highlights token efficiency: Astra uses 22x fewer input tokens than Sonnet 5.5 and 5x fewer than Opus 5.5, while Sol beats both at roughly a tenth of Sonnet's cost — suggesting scientific research is starting to scale with compute.
Related event: GPT-6 Astra tops new Vals Science Index at far lower cost(2 posts)→
More from Models
- DeepSeek V4.1 Flash 1017 reportedly scores 42.3 on Terminal-Bench 4.0, beating GLM-5.3 but only with its own harness — teortaxesTex · 2026-10-19
- Viral theory claims RSI-capable models can't be open-sourced, DeepSeek dates are frozen snapshots — teortaxesTex · 2026-10-19
- Claude Is Today's Busiest Bridge in AI Discourse Graph as Anthropic Brand Buzz Cools 17% — YvesMulkers · 2026-10-19
- Programmer's 'insane prompting' is just begging Claude to fix code before going home — _jaydeepkarale · 2026-10-19
- Gemini Argon 4 hands-on: solid at Go debugging, a second choice after Opus 5.5 — DanielLockyer · 2026-10-19
- Two German courts now hold Google liable for false AI Overviews content — emmanuelvivier · 2026-10-19