Unverified50% confidenceFactExact time
根据Commander Code官方内部评测数据，Kimi K2在高难度工具依赖任务中获得10次中5次的胜率，几乎追平Claude Opus
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026
Sources
开源模型写代码总报错？问题可能出在编程框架上
bilibili小牛AI_XNAI6/11/2026
Related Claims
Unverified根据Commander Code官方内部评测数据，DeepSeek V4 Pro在高难度工具依赖任务中10次有6次击败了Claude Opus71% similarUnverifiedKimi K2在测试中表现出较快的执行速度，很快显示Complete状态68% similarUnverifiedClaude Opus在测试中7分30秒一次成功完成任务，动画流畅、功能完整，排名第一63% similarUnverifiedKimi K2.7 Code采用384个专家的MoE架构，每个token选择8个专家加1个共享专家，实际计算量约为稠密模型的2.3%62% similarUnverified相比Kimi K2.5，Kimi K2.6的API处理能力更强，运行更稳定，任务完成率显著提升62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/49111API
curl https://kongchang.com/api/v1/knowledge/claims/49111MCP
get_claim(id=49111)