Kimi

Kimi, from Chinese startup Moonshot AI, offers one of the largest context windows around — up to a million tokens — letting it read entire books or massive codebases in a single prompt, with a reasoning line that's scored well on international math and coding benchmarks against OpenAI and DeepSeek's models. Moonshot also released Kimi K2 as an open-weight agentic model in 2025, which got real attention as another example of a Chinese lab shipping frontier-competitive open models.