Kimi K3 Is Competitive with Fable; Kimi K3 and Fable Is SoTA

Kimi K3 and Fable are competitive with each other and together achieve state-of-the-art performance through task-level routing, according to a study comparing the models on approximately 1,030 agentic tasks. The study found that Kimi K3 was selected by an oracle router for 72-96% of tasks, indicating near-perfect routing potential. While the models show similar overall accuracy across benchmarks, they excel in different task types, with Kimi K3 stronger in symbolic math and terminal work and Fable ahead in multi-language coding and visualization. Despite comparable quality, Kimi K3 operates at a fraction of the cost due to efficient token usage and prompt caching. Routing tasks to the better-performing model yields higher quality than either model alone at a cost close to the more efficient option. The results suggest that the best AI performance now comes from combining specialized models rather than relying on a single provider.