Kimi K3 vs Fable 5: same code, a third of the price, four times slower
The author ran Moonshot's Kimi K3 and Anthropic's Fable 5 through three real coding tasks in the Rust fd repo: a bug fix, a refactor, and a new flag. The prompts were identical; only the model and CLI changed, Kimi Code 0.27.0 against Claude Code 2.1.212.
The output matched. On the bug, both deleted the same line in src/main.rs, the diffs came out byte for byte, and all 70 tests passed. The refactor produced a config_builder.rs module and cleared 264 tests in both runs, and both shipped a working --count flag.
On API pricing, Kimi K3 runs $3 per million input tokens and $15 per million output; Fable 5 is $10 and $50. Across the three tasks Kimi cost $2.13 to Fable's $5.98, but burned 3.3 million tokens against 2.4 million. On the clock Fable finished in 6 minutes 49 seconds; Kimi took about 28 minutes 18 seconds.
Related stories
- Claude Fable 5 vs GPT-5.6 Sol: a $100 music video bake-off
- Claude Sonnet 5 increases performance at a higher cost per task
- Meta shipped three benchmark charts with Muse Code. Claude Opus 5 wins all three.
- Qwen 3.8 and Claude Opus 5 show why raw benchmark scores don't predict the bill
- GLM-5.2 (max) matches Claude Opus 4.8 on Harvey LAB-AA
- GLM-5.2 vs. Claude Opus 4.8
Comments
No comments yet. Be the first.
Join the conversation
Sign in with Google to leave a comment. Your name and avatar come from your Google profile, and the comment appears after moderation.