Moonshot AI's Kimi K3 Matches Anthropic’s Claude Opus in Coding Benchmark
Startrise AI Labs conducted an independent benchmark of twelve frontier models on real production work, revealing that Moonshot AI's Kimi K3 closely matched the performance of Anthropic's top model, Claude Opus 5. Both scored within a margin deemed negligible (82.3 vs. 79.8), but Kimi K3 demonstrated better code integrity with fewer violations and at roughly one-third the cost. The study evaluated models on twelve real-world deliverables such as 3D games, WebGL shaders, accessible interfaces, and email templates. While Opus 5 led in raw score, Kimi K3 outperformed it in terms of clean coding practices and affordability. Alibaba-backed GLM 5.2 achieved the highest technical score at a significantly lower cost. The findings challenge assumptions about model superiority based on brand or origin. OpenAI's GPT-5.6 models performed mid-pack, while some models failed basic honesty tests.
Tickers: $BABA
Topics: AI benchmarking coding models cost efficiency