Kimi 发了新编程模型 K2.7 Code，推理更省 token，基准提升明显，而且在 Agent Arena 上开源模型里排第6，值得一试。
Kimi K2.7 Code 在 Agent Arena 排行榜上总体排名第19，在开源模型中排第6。该模型在 Kimi Code Bench v2 上比 K2.6 提升21.8%，在 Program Bench 上提升11.0%，在 MLS Bench Lite 上提升31.5%。推理 token 使用量降低30%，减少了过度思考。长程编码任务指令遵循和完成率均有提升。目前通过 Kimi API 和 Kimi Code 可用。
Kimi K2.7 Code by @Kimi_Moonshot ranks #19 overall on the new Agent Arena leaderboard, and #6 among ...
Kimi K2.7 Code by @Kimi_Moonshot ranks #19 overall on the new Agent Arena leaderboard, and #6 among open models. In Agent Arena, we measure models on millions of real-world, long-horizon agentic tasks from a global community of users. Models can access web search, filesystem, and terminal tools to complete complex workflows. The leaderboard measures model performance on outcomes relative to the average model using a causal tracing methodology. Kimi K2.7 Code's strongest signal is confirmed task success, while bash capabilities and tool hallucination hold stable. The tradeoff is steerability, which regresses sharply compared to K2.6 (-12.25% vs. -2.82%). Note the wide confidence intervals, since the scores are still stabilizing. See thread for details on how Kimi K2.7 Code performs across 5 different signals. Kimi.ai @Kimi_Moonshot 🌘 Kimi-K2.7-Code, our latest coding model, is now released and open-sourced! 🔷 Improved coding & agent performance over K2.6: +21.8% on Kimi Code Bench v2, +11.0% on Program Bench, and +31.5% on MLS Bench Lite. 🔷 Reasoning efficiency: Less overthinking, with 30% lower reasoning-token usage compared to K2.6. 🔷 Long-horizon coding: Improved instruction following, higher end-to-end coding task success rates. ⚡️ 6x High-Speed Mode coming soon! 🔌 Available today via Kimi API and Kimi Code. 🔗 Kimi kimi.com/code JKyGCY � platform.moonshot.ai kbOwCN4 🔗 View Quoted Tweet 💬 5 🔄 2 ❤️ 30 👀 2148 📊 7 ⚡