Devin Fusion Achieves 60% Cost Reduction on FrontierCode Performance

Cognition's Devin Fusion delivers a cost score of $1.35 on the FrontierCode benchmark, representing a 60% reduction compared to traditional models. This multi-model harness integrates various AI models, including Fable 5, Opus 5, and GPT-5.6 Sol, achieving significant cost efficiencies. For instance, the cost scores for Fable 5, Opus 5, and GPT-5.6 Sol are $10.53, $3.51, and $3.41, respectively.

Devin Fusion has been tested alongside Fable 5 prior to its access suspension on June 12, 2026, due to a US government directive. The cost improvements with Devin Fusion include a 41% reduction compared to using Fable 5 alone and up to 60% savings when compared to Opus and GPT-5.5.

Cognition reports that 88% of merged pull requests (PRs) were driven by the automated Fusion router, indicating a substantial impact on development efficiency. Specific examples of cost savings include 62% on test run costs, 32% on mechanical removal tasks, and 25% on integrating WebSocket MCP transport.

Cognition emphasizes that the largest efficiency gains are observed during implementation-focused sessions, suggesting that practitioners can expect enhanced performance and reduced costs in their AI deployments with Devin Fusion.