Dispatch
A measured Astra-vs-Sol agent run: medium reasoning was cheapest and fastest, high was neither
A gateway vendor benchmarked a three-phase autonomous coding pipeline using Galley inside the Codex CLI. Astra Medium finished at $25.67 total, 19% under Sol High ($31.79) and 31% under Astra High ($37.23), and completed implementation in 31 minutes against Sol High's 52. The token gap is the interesting part: Sol High burned 37.8M input tokens across 238 requests from repeated context reloading, while Astra Medium finished the same phase on 11.1M input tokens over 80 requests. Single-source and published by a router with an obvious interest in the conclusion, so treat the ranking as a hypothesis to re-run on your own stack.
Source
↳ Follow the thread