Devin Fusion's dual-model setup now runs in Devin Desktop and CLI

Your move: Install Devin CLI to pick your own lead and sidekick models

Cognition brought Fusion, its dual-model setup for Devin, to Devin Desktop and Devin CLI on September 11, 2026. Fusion pairs a frontier "lead" model that handles planning and review with a cheaper "sidekick" model that does the execution, running the two in parallel. It was previously only available on Cognition's cloud platform; now it can run locally through Desktop or CLI, with users choosing their own lead and sidekick models. Cognition recommends pairing Fable 5.1 as the lead with SWE-2 as the sidekick, and reports the combination costs 36% less than using Claude Code with Fable 5.1, and 39% less than Codex with Astra, while matching or beating their coding benchmark scores.

Why it matters

If you run Devin Desktop or CLI locally, switching on Fusion with the recommended model pairing gets you comparable or better coding results at meaningfully lower cost. Worth trying if you're watching your API spend on agentic coding work.

Source: Cognition blog ↗

Published by Agent Log. The summary reports the linked source; “Why it matters” and the verdict are Agent Log's interpretation.

More Devin Desktop updates →