Astra uses about a third of Sol's tokens on coding agent tasks, Artificial Analysis says

Rajath Gowda @buildwithrajathOriginally posted 2026.09.05

Open original on X

🚨 GPT-6 Astra might make a lot of coding models look bloated. Artificial Analysis says Astra uses ~3x fewer tokens than GPT-5.6 Sol on coding-agent tasks. And it still scores 67 on the Coding Agent Index. The crazy part? At max effort, Astra costs roughly the same per task as Sol - while scoring higher. - Fewer tokens. - Same cost. - Better output. As someone building with AI, that’s way more interesting to me than another benchmark score. The real advantage of Astra might be how much work it gets done per token.