An independent benchmark posted on X found Devin Fusion achieving a 62 score on the Coding Agent Index v1.5, matching the performance of Claude Fable 5.1 while offering lower cost options.
The community user @ArtificialAnlys reported testing two configurations of Devin Fusion that combine a frontier lead model from Anthropic or OpenAI with Cognition’s SWE‑2 (medium) sidekick. The Claude Fable 5.1 (xhigh) + SWE‑2 setup scored 62, whereas the GPT‑6 Astra (xhigh) + SWE‑2 configuration scored 59, was 43% less expensive and completed tasks 31% faster.
Cognition’s documentation on Fusion states the feature is “the most efficient frontier harness for Fable and Astra, up to 39% more efficient compared to other model harnesses across major coding benchmarks” and notes that Fusion is now available in Devin Desktop and CLI.
The benchmark marks the first inclusion of a multi‑model coding agent on the Artificial Analysis Coding Agent Index, according to the community post.
Sources
Fan coverage from Devin Central — not an official Cognition announcement. Devin is a trademark of Cognition.