Kimi K3 Matches Fable 5 on 3 Coding Tasks at One-Third Cost, Taking 4 Times Longer
Updated
Updated · The New Stack · Jul 20
Kimi K3 Matches Fable 5 on 3 Coding Tasks at One-Third Cost, Taking 4 Times Longer
3 articles · Updated · The New Stack · Jul 20
Summary
Across three real coding jobs in the fd Rust codebase, Moonshot AI’s Kimi K3 produced the same end results as Anthropic’s Fable 5 for $2.13 versus $5.98, but needed about 28 minutes against 6 minutes 49 seconds.
The gap came from pricing rather than efficiency: Kimi’s API is priced at $3 per million input tokens and $15 per million output, versus Fable 5 at $10 and $50, even though Kimi consumed more tokens overall—3.3 million versus 2.4 million.
On the tasks themselves, both models delivered an identical bug fix, passed all 264 refactor tests, and built a working --count feature; Kimi was more exhaustive in validation during the refactor, while Fable also updated zsh completions in the feature task.
The test used identical prompts and native CLIs because Kimi K3 is not yet integrated into coding tools like Cursor, leaving model behavior and toolchains as the main variables.
The results suggest Kimi K3 has credible coding quality and aggressive pricing, but its much slower turnaround remains the main obstacle as Moonshot tries to challenge established coding models.