Claude Sonnet 5.5 Comes to Devin, Beating Fable 5.1 on FrontierCode at a Lower Cost

Cognition made Claude Sonnet 5.5 available as a model option inside Devin on September 28, 2026, the same day it announced a broad cost-efficiency update for the platform.

Details

• Benchmark result: Claude Sonnet 5.5 scores 64.4% on FrontierCode 1.1, Cognition’s own coding benchmark, surpassing Fable 5.1 even at Fable’s extra-high reasoning effort setting
• Speed and cost: Cognition reports more than 30% faster output generation and up to 30% lower cost per task compared with the earlier Sonnet 5, while token pricing itself is unchanged
• Access: Sonnet 5.5 is selectable as a model in Devin sessions through Devin Desktop and the Devin CLI, alongside the FrontierCode leaderboard where Cognition publishes comparative benchmark results
• Context: the addition follows Cognition’s pattern of adding frontier models to Devin quickly after their release, having already added Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and Grok 4.7 earlier in September

What happened next

Pairing the Sonnet 5.5 rollout with the same-day announcement that Devin got up to 40% cheaper across the board suggests Cognition is treating each new frontier model release as an opportunity to re-tune Devin’s cost-performance curve rather than simply bolting on another option. With Sonnet 5.5 now beating Fable 5.1 on Cognition’s own leaderboard at a lower cost than its predecessor, Devin’s default model recommendations are likely to shift toward Sonnet 5.5 for teams optimizing for both benchmark quality and per-task spend.