Devin is now up to 40% more cost-efficient
By Steven Van ·
Fusion and Normal mode cost 30-40% less, Ultra 15-20% less and Devin Review up to 70% less, with Fusion scoring 68.8 on FrontierCode 1.1.
Devin's per-task cost has dropped across its modes. Devin is now 30-40% cheaper in Fusion and Normal mode, 15-20% cheaper in Ultra, and up to 70% cheaper in Devin Review, according to the vendor's announcement. The company says intelligence has been maintained or improved in every mode.
Where the savings come from
Devin attributes the reductions to newer models, including its own SWE-2, and to changes in its Cloud harness. It says it picks models per job:
- Fusion pairs a capable lead model with a more cost-effective sidekick.
- Devin Review chooses models for analyzing diffs, finding bugs and categorizing changes.
- Normal and Ultra choose models for planning, testing, debugging, migrations and similar work.
The post describes Opus 5.5 and GPT-6 Sol as stronger and price-performant, GPT-6 Astra as good at computer use, and GPT-6 Luna as substantially cheaper for supporting tasks.
Harness changes
Devin now batches several shell commands into one call and runs independent tool calls in parallel. A model can request formatting, linting and tests together, then reason about the results at once. In the vendor's illustrative example, that takes two turns instead of four and sends 49% fewer tokens.
The harness also reuses cached context more. In another illustrative session, caching cut tokens processed from scratch by 71%. The vendor notes that cache reads are getting cheaper: Opus 5.5's cost 60% less than Opus 5's, and GPT-6 Sol and Luna both halve cache-read prices compared with their 5.6 predecessors.
Benchmark result
Devin Fusion scores 68.8 on FrontierCode 1.1 Extended, at an average of $0.60 per task. Devin lists Opus 5.5 (high) at 65.2 and $0.90, Fable 5.1 (medium) at 63.6 and $2.68, GPT-6 Astra (high) at 63.1 and $2.62, SWE-2 (high) at 60.1 and $0.64, and GPT-6 Sol (high) at 59.6 and $0.87.
The more efficient Devin is available now at devin.ai.
