Astra hits 74% on DeepSWE with 30k tokens, half the cost steps of GPT-5.6 Sol

haider1 · x · 2026-09-05

A comparison on the DeepSWE benchmark shows Astra scoring 74% using only 30k output tokens across 29 steps ($6.52), while GPT-5.6 Sol reaches a similar 73% but needs 60k tokens and 61 steps ($6.46) — roughly double the token and step count. The poster marvels at Astra's efficiency, especially given Sol launched just a few months ago.

Original post →

More from Models

Models channel →