Kimi K2.6

Moonshot's open-weight model built for long-running agentic and tool-use tasks

Free Reasoning
Visit Product Page →
256000 context tokens
1T (32B active) parameters
Modified MIT license
Apr 2026 released

Kimi K2.6 is Moonshot AI’s open-weight flagship model, released in April 2026 as the successor to the original Kimi K2. It is a 1-trillion-parameter mixture-of-experts model that activates about 32 billion parameters per token, ships natively in INT4 for cheaper inference, and handles a 262,144-token context window. The model is built for long-running agentic work: Moonshot designed it around multi-agent orchestration, coding sessions that run for hours without human intervention, and native multimodal input rather than bolted-on vision. Independent evaluations from Artificial Analysis rank it as the strongest open-weight model available, putting it within a few points of closed frontier systems from OpenAI and Anthropic on tasks like SWE-Bench Pro and Humanity’s Last Exam, while charging a fraction of their per-token price. Moonshot releases the weights under a modified MIT license, so anyone can download and self-host the model rather than relying on Moonshot’s own API.