Woofun AI data shows that Datacurve has integrated Kimi K3 into the DeepSWE long-range software engineering benchmark, which evaluates a model's ability to independently complete complex, real-world development tasks. Kimi K3 recorded a task success rate of 69%, placing it just one percentage point behind Claude Fable 5's 70% and four points below GPT-5.6 Sol's 73%. Datacurve identified Kimi K3 as the first open-weight model to reach the performance level of cutting-edge closed-source models on this specific benchmark.

The DeepSWE suite comprises 113 original tasks drawn from 91 active open-source projects across five programming languages, with all models utilizing an identical toolset to minimize training data contamination. In terms of efficiency, Kimi K3 incurred an average cost of $4.65 per task, significantly lower than GPT-5.6 Sol's $8.39 and Claude Fable 5's $21.63.