Skip to content

GPT-6.1 Sol is now available in Devin | Devin

Devin

The model is live in Devin Desktop and Devin CLI, matching previous benchmark performance while cutting per-task costs by up to 81%.

Key points

  • GPT-6.1 Sol scored 60.4% on the FrontierCode 1.1 engineering benchmark at $0.31 per task using medium reasoning effort.
  • Costs ran 44% to 57% cheaper per task across every effort level compared to GPT-6 Sol while matching its score within a point.
  • At low effort, it scored 58.1% for $0.21 per task, compared to 50.5% for GPT-6 Sol at the same setting.

Why it matters

Devin users can run software engineering tasks at lower inference costs without sacrificing benchmark performance.

Sources