saved
Introducing Fusion in Devin CLI
Hraness cites a source capture. The source author remains the source.
Makers of Devin, the first AI software engineer. We are an applied AI lab building end-to-end software agents. Join us: https://t.co/4Ss9hvpjRG
Introducing Fusion in Devin CLI
The most efficient frontier harness for Fable & Astra; 39% cheaper across coding benchmarks.
Pick your favorite model for planning and a cost-effective model for execution.
We partnered with @ArtificialAnlys and @ValsAI to evaluate Fusion across several coding agent benchmarks. The cost savings hold across the board while maintaining frontier performance.
Unlike plain model routing, Fusion keeps frontier intelligence in charge. We don’t rely on one initial routing decision. The lead reviews work, flags problems, and can take control back when the sidekick is out of its depth.
Try it out in Devin CLI: devin.ai/cli
Read more in our blog:
Chart. Artificial Analysis Coding Agent Index v1.5. Score (higher is better): Claude Code Fable 5.1 (max) 62.2; Devin Fusion Fable 5.1 + SWE-2 61.7; Codex Astra (max) 61.6; Claude Code Opus 5 (max) 60; Devin Fusion Astra + SWE-2 58.9; Muse Code Muse Spark 1.3 (max) 54; Opencode GLM-5.3 54; Kimi Code CLI Kimi K3 52; Grok Build Grok 4.6 (xhigh) 47; Claude Code Qwen3.8 Max 43; Codex DeepSeek V4 43; Antigravity SDK Gemini 3.8 Flash 42. Cost (lower is better): Claude Code Fable 5.1 versus Devin Fusion Fable 5.1 + SWE-2 −36%; Codex Astra versus Devin Fusion Astra + SWE-2 −39%.
Table. Benchmark | Fable 5.1 | Fusion (Fable 5.1 + SWE-2) | Astra | Fusion (Astra + SWE-2). DeepSWE 1.1: 64.3 $14.63 | 63.1 $7.88 (−46%) | 67.6 $7.88 | 67.3 $4.69 (−40%). Terminal-Bench 4: 57.6 $17.46 | 56.1 $13.37 (−23%) | 55.6 $10.08 | 50.0 $6.06 (−40%). SWE-Atlas QnA: 64.8 $7.57 | 65.9 $5.00 (−34%) | 61.8 $5.72 | 59.4 $3.59 (−37%). Vals Code Migration: 54.6 $70.97 | 57.3 $42.00 (−41%) | 67.7 $44.36 | 61.3 $35.51 (−20%). FrontierCode 1.1 (Extended): 63.6 $2.68 | 63.5 $1.67 (−38%) | 63.1 $2.62 | 63.4 $2.34 (−11%).
Diagram. Main Agent and Sidekick. Main Agent: User task → File snippets → Plan → Review code → Request edits → Final code. Sidekick: Code exploration; Write code / Write tests / Fix lint; Fix bugs. Flow: User task to Code exploration to File snippets; Plan to Write code / Write tests / Fix lint to Review code; Request edits to Fix bugs to Final code.
Terminal. Devin CLI v2026.9.2 · Teams · 1.0M token context. > /model. Model set to Fusion. Lead Claude Fable 5.1 · Sidekick SWE-2 High. > Upgrade us to Next.js 16 and fix whatever breaks. Run npx @next/codemod@latest upgrade 16. next 15.5.2 -> 16.0.0, 38 files changed. Fix async params in 14 route handlers.
