Claude Opus 5 Now Available in Devin
WindsurfDevin added support for Claude Opus 5 across Devin Desktop and Devin CLI, with Cloud access to follow as part of its model mix. On Cognition's FrontierCode 1.1 benchmark, Opus 5 scored 63.6, trailing only Claude Fable 5 (64.9) and beating GPT-5.6 Sol, GPT-5.5, and Claude Sonnet 5. Within Devin specifically, the model showed particular strength on debugging and root-cause analysis, favoring targeted fixes over large refactors and closely following existing repo conventions.
Featured Video
A video we selected to help illustrate this changelog
Key Takeaways
- Devin now runs Claude Opus 5 across Desktop, CLI, and soon Cloud's mode mix, marking the newest Anthropic flagship model's first appearance inside the coding agent.
- FrontierCode 1.1 places Opus 5 second only to Claude Fable 5, scoring 63.6 versus Fable 5's 64.9, while beating GPT-5.6 Sol, GPT-5.5, and Claude Sonnet 5.
- Early reviewers call Opus 5 an unusually strong debugger, with YouTube first-look videos highlighting its ability to trace root causes rather than just patch symptoms.
- The model favors targeted fixes over sweeping refactors, a trait multiple reviewers flagged as making it safer for production codebases than more aggressive rewrite-happy models.
- Independent testers describe Opus 5 as "brilliant but annoying," noting a demanding, opinionated working style even as it outperforms rivals in blind benchmark comparisons.
- Pricing lands at parity with Opus 4.8 ($5/$25 per million input/output tokens), giving Devin users near-Fable-level coding quality without the top-tier price tag.
Sources & Mentions
5 external resources covering this update
Devin Adds Claude Opus 5 Across Desktop, CLI, and Cloud
Devin now runs Claude Opus 5, Anthropic's newest flagship model, inside Devin Desktop and Devin CLI, with the model set to join Devin Cloud's mode mix shortly after. The rollout gives Devin users access to a model Cognition describes as approaching Fable-level intelligence at roughly half the cost, making it one of the more cost-efficient frontier options available inside the platform.
Benchmark Performance
On FrontierCode 1.1 Extended, Cognition's proprietary benchmark that grades models on real-world engineering tasks based on code quality and mergeability, Claude Opus 5 scored 63.6 out of 100. That places it second only to Claude Fable 5's 64.9, and ahead of GPT-5.6 Sol (60.6), Claude Opus 4.8 (59.6), GPT-5.5 (56.7), Claude Sonnet 5 (56.2), GPT-5.6 Terra (55.8), and SWE-1.7 (54.6). The benchmark's scoring is a weighted aggregate of rubric items, and solutions that fail blocking criteria receive a score of zero, making the results a reasonably strict measure of production-readiness rather than raw capability.
Debugging Strength and Coding Style
Cognition's evaluations found Claude Opus 5 particularly strong on difficult debugging and root-cause analysis tasks. Several behavioral patterns stood out during testing: the model favors targeted, in-place bug fixes rather than sweeping refactors, adheres closely to existing repository conventions when building new features and writing tests, and follows detailed specifications with a high degree of completeness. Independent reviewers echoed these findings, with early hands-on videos and newsletter write-ups describing Opus 5 as unusually judgment-driven, reasoning through problems before writing code and catching logical faults during planning rather than after the fact. Some reviewers noted the model's demanding, opinionated working style even as it outperformed rivals in blind side-by-side comparisons.
Pricing and Access
Claude Opus 5 is priced at parity with Opus 4.8, at $5 per million input tokens and $25 per million output tokens, and ships with a 1M-token context window. For Devin users, this means access to near-Fable-tier coding performance without paying Fable-level prices. The model is available now through Devin Desktop and Devin CLI, with Devin Cloud access rolling out as part of its broader mode selection.