DeepSWE Eval: Kimi K3 Matches Claude Fable 5 at 35% of the Cost

togethercompute · x · 2026-07-22

Together AI conducted a model comparison for software engineering tasks based on the DeepSWE benchmark.

Results show that Kimi K3 achieves the same performance level as Claude Fable 5 at approximately 35% of the price. Furthermore, at higher pass@k settings (allowing more generation attempts), Kimi K3 actually pulls ahead of the competition.

Related event: Kimi K3 Matches Fable 5 in SWE Tasks at a Third of the Cost(7 posts)→

Original post →

More from Models

Models channel →