Open-Source Coding Model Race Heats Up

Kimi K3, GLM-5.2, Xiaomi MiMo-V2.5-Pro, DeepSeek V4 Pro, and now Kimi K2.7 Code compete on benchmarks like 10xBench, Frontend Code Arena, and SWE-bench. GLM-5.2 leads 10xBench (8.5) but is easily jailbroken. Kimi K2.7 Code offers 256K context and 30% fewer reasoning tokens but raises cost concerns at 78¢/prompt.