Cognition slashes Devin usage costs up to 70% in Review
Cognition published an efficiency update for Devin that lowers usage costs across modes, with Devin Review seeing reductions of up to 70%. Windsurf's blog attributes the savings to newer models and a more efficient harness, including work related to SWE-2. Fusion and Normal modes are said to be 30-40% cheaper, while Ultra is 15-20% less expensive. The company also cites Devin Fusion scoring 68.8 on FrontierCode 1.1 Extended at about $0.60 per task. Cognition states the rollout is live on devin.ai. For operators comparing autonomous coding agents, the shift is materially cheaper review workloads, though cost claims rest on vendor reporting rather than independent verification.
Cognition slashes Devin usage costs up to 70% in Review
Cognition said Devin is now 30-40% cheaper in Fusion and Normal mode, 15-20% cheaper in Ultra, and up to 70% cheaper in Devin Review after model and harness work including SWE-2. The post also reports Devin Fusion at 68.8 on FrontierCode 1.1 Extended at about $0.60 per task and states the efficiency update is live at devin.ai.
Key takeaway
Devin’s steepest advertised savings are in Review (up to 70%), making review-heavy agent pipelines the main budget lever to reassess first.
What happened
Cognition said Devin is now 30-40% cheaper in Fusion and Normal mode, 15-20% cheaper in Ultra, and up to 70% cheaper in Devin Review after model and harness work including SWE-2, according to reporting summarized in the evidence packet.
The same announcement, carried on the Windsurf blog, also reports Devin Fusion at 68.8 on FrontierCode 1.1 Extended at about $0.60 per task and states the efficiency update is live at devin.ai.
Evidence
Devin Review usage can be up to 70% cheaper after the update.
Windsurf Blog · attributed
up to 70% cheaper in Devin Review, thanks to the latest models and a more efficient harness.
Fusion and Normal modes are 30-40% cheaper; Ultra is 15-20% cheaper.
Windsurf Blog · attributed
Devin is now 30-40% cheaper in Fusion and Normal mode, 15-20% cheaper in Ultra
Devin Fusion scored 68.8 on FrontierCode 1.1 Extended at about $0.60 per task.
Windsurf Blog · attributed
Devin Fusion at 68.8 on FrontierCode 1.1 Extended at about $0.60 per task
The efficiency update is live on devin.ai.
Windsurf Blog · attributed
states the efficiency update is live at devin.ai.
Why it matters
If the published per-mode discounts hold in real invoices, autonomous review and fusion workflows get meaningfully lower run-rate costs without waiting on a future release.
Limits and uncertainties
Cost and benchmark figures come from Cognition’s announcement via the Windsurf blog; the packet does not cite independent billing audits or third-party verification.
The packet does not specify how savings vary by task type, team plan, or whether quality or latency changed alongside efficiency.
Practical implications
Teams on Devin Review should compare pre- and post-update usage bills once the live devin.ai pricing reflects the stated cuts.
Buyers evaluating agents on economics should treat the FrontierCode 1.1 Extended score and $0.60-per-task figure as vendor-reported datapoints, not validated benchmarks.
What to watch
Whether devin.ai invoices and in-product meters show the advertised 30-40%, 15-20%, and up-to-70% mode-specific reductions.
Follow-on Cognition or Windsurf posts clarifying SWE-2 harness changes and any tradeoffs in Ultra versus Review quality.
