China’s Kimi K3 Outperforms GPT 5.6 Sol in SWE Marathon

China’s Kimi K3 reportedly outperformed GPT 5.6 Sol and other leading U.S. models on reasoning benchmarks, including SWE Marathon.

It uses a Mixture of Experts architecture to improve performance while reducing hardware requirements and cost.

The result shows Chinese AI developers competing effectively under chip restrictions through efficient architectures and software optimization.

That could reshape global AI competition and speed adoption of cost-efficient models.