Kimi k1.5: the Chinese competitor to OpenAI o1 with 128k context and long-thinking
In one sentence Moonshot AI releases Kimi k1.5, a reasoning model with 128k context and RL-trained long chain-of-thought that matches OpenAI o1 on AIME and MATH-500, with a user-controllable 'long-thinking' mode.
By January 2025, OpenAI had already launched o1, a model capable of "thinking before answering" on complex mathematical and logical problems. Moonshot AI, a Chinese company, demonstrated it could do the same — and possibly better.
Kimi k1.5 introduces a "long-thinking" mode: instead of immediately giving an answer, the model can take much more time to reason step by step, exactly like a student working through a difficult problem writing out every step on paper. And importantly, the user can control how long the model "thinks" before responding.
On difficult university-level math benchmarks like AIME and MATH-500, Kimi k1.5 achieves results comparable to OpenAI o1 — a remarkable result for a lab outside the United States. And it can work with very long texts: up to 128,000 words of context, the equivalent of an entire book.
Companies
Moonshot AI
Tools
—
Tags
Sources