Kimi K3.1 Pro lands in Devin Desktop, CLI

K3.1 Pro’s Devin debut makes a strong open-weight coding model easier to use, but the improvement is effort-sensitive rather than universally dominant.

• –Max-effort performance rises 1.9 points over Kimi K3 on FrontierCode 2.0
• –At high effort, its 17.3 score trails Kimi K3’s reported 17.7 best
• –Devin users get access across both desktop and terminal workflows without separate model plumbing
• –Its 1M-token context and stronger long-horizon coding target complex, multi-file engineering tasks
• –FrontierCode 2.0 is Cognition’s private benchmark, so independent evaluations will matter before treating the gain as definitive