Fireworks AI benchmarks Kimi K3 against Fable 5

In a study of ~1,000 agentic tasks, Kimi K3 achieved a 92.4% score on SWE benchmarks compared to 92.6% for Fable 5. The analysis suggests K3 is suitable for 72-96% of tasks in an oracle routing setup, with K3 showing superior performance in symbolic math and development tasks.