BREAKING
Kimi K2.7 Code: 19th on Agent Arena
0
th
Overall rank
0
th
Among open models
0
T
Total params
0
B
Active params
0
K
Context length
K2.7 Code vs K2.6 benchmarks
Code Bench v2
62
Program Bench
53.6
MLS Bench Lite
35.1
MCP Atlas
76
MCP Mark
81.1
Strengths and open questions
Strengths
Pros
●
Cuts reasoning tokens ~30%
●
Beats Claude and GPT on tool use
●
Strong long-context following
Open questions
Watch
●
Steerability dropped vs K2.6
●
No third-party SWE-Bench check
●
Wide confidence intervals
Verdict awaits independent review
AI NEWS BLITZ
Moonshot AI's new coding model just landed a top ranking on the Agent Arena leaderboard.