Devin gets 30–70% cheaper as Fusion tops FrontierCode 1.1
Savings come from SWE-2 routing and prompt-cache gains, landing as Fusion retakes the top coding benchmark — recheck your usage caps today.
Copy markdown
Devin's bill drops 30–70%, mode by mode
Cognition made Devin 30–40% cheaper in Fusion and Normal, 15–20% in Ultra, and up to 70% in Devin Review, with no capability regression. Fusion runs now average about $0.60 per task — the same work you shipped last week costs less today, so revisit any spend caps you set around the old rates.
Fusion takes FrontierCode 1.1, on Cognition's own SWE-2
In the same release, Devin Fusion hit first on FrontierCode 1.1 Extended (68.8), routing across Cognition's in-house SWE-2 alongside Opus 5.5 and GPT-6 variants. Cheaper and higher-scoring at once makes it worth re-benchmarking Fusion against whatever agent you run now — before today's DevDay reshuffles the leaderboard.