Kimi K3 matches Claude Opus 4.8 on benchmarks at 40% lower cost

Kimi K3, an open-weight model, ranks fourth on the Artificial Analysis Intelligence Index, statistically tied with Claude Opus 4.8 and GPT-5.5. It costs 40% less per token and leads in frontend code generation, but lacks SWE-bench results and ecosystem maturity. The choice depends on workload: K3 for cost and open weights, Opus for agentic reliability.