Kimi K2.6 Leads 5 of 8 Major Coding Benchmarks — Only Open-Weight Model in Top Tier

Kimi Blog·high signal

Moonshot AI's Kimi K2.6 now leads on 5 of 8 major agentic coding benchmarks while remaining the only open-weight model in the comparison. On SWE-Bench Pro it scores 58.6% vs GPT-5.4's 57.7% and Claude Opus 4.6's 53.4%. Also leads on HLE with tools (54.0%) and DeepSearchQA (92.5% F1). For builders: K2.6 is now the strongest choice for agentic coding workflows, saving up to 88% vs proprietary models.