Kimi K3 in OpenCode: What One Finished Coding Task Actually Cost

Short summary

A controlled test of Kimi K3 completing a leap-year bug fix through OpenCode cost $0.19 across 28,707 tokens and 84 seconds, with all tests passing. The breakdown reveals cached-read tokens dominated usage, highlighting why per-token pricing alone misleads coding agent cost estimates. The author argues task-completion metrics—steps, tool calls, tokens, and billed usage—matter more than benchmark scores.

• •Kimi K3 fixed a leap-year bug via OpenCode for $0.19 in 84 seconds with all tests passing
• •28,707 total tokens were 80% cached-read tokens, showing real agent cost differs from per-token pricing
• •Author proposes task-completion cost as a better evaluation metric than benchmarks or token prices alone

Generated with AI, which can make mistakes.