Devin Fusion benchmarks: frontier lead model plus cheap sidekick cuts coding agent costs

ArtificialAnlys · x · 2026-09-12

Artificial Analysis independently benchmarked Devin Fusion, Cognition's newly released multi-model coding agent and the first of its kind on their Coding Agent Index.

Fusion pairs a frontier lead model (Claude Fable 5.1 or GPT-6 Astra, both xhigh) with a cost-efficient SWE-2 (medium) sidekick. The Fable config scores 62 on the Index v1.5, while the Astra config scores 59 but is 43% cheaper and 31% faster — retaining frontier-level performance at lower cost.

Original post →

More from coding & agent

coding & agent channel →