Kimi K3 vs Claude Fable 5 on DeepSWE: Cost and Coding

Together AI ran 452 DeepSWE rollouts on the two models: Claude Fable 5 edges pass@1 by 1.4 points, but Kimi K3 wins pass@4 and delivers 2.8x the solves per dollar. K3 lands on Together at ~35% of Fable's price; a separate run found Kimi K3 Max matches GPT 5.6 Sol Max at ~55% of the cost.