Kimi K3 matches Claude Fable 5 on agentic coding benchmarks

Kimi K3, Moonshot AI's open-weight model built on Kimi K2's MoE architecture (over 1T total params, ~32B active), matches Claude Fable 5 on coding benchmarks at Sonnet-level pricing. The comparison weighs tool use accuracy, multi-step reasoning, context handling, and deployment cost for agentic workflows.