Moonshot AI Opens Source Its Strongest Model Kimi K2.6: Code Capability Matches GPT-5.4, Agent Cluster Supports 300 Parallel Subtasks

Developers often struggle with low efficiency in complex coding tasks. Moonshot AI has released and open-sourced its latest model Kimi K2.6, whose code capability matches GPT-5.4, and its Agent cluster supports parallel processing of 300 subtasks.

The model performs leadingly in tests like SWE-Bench Pro, with significantly improved long-term coding ability—it can code continuously for 13 hours and modify over 4000 lines of code. The Agent cluster can deliver multi-product end-to-end results, such as generating quantitative strategies for semiconductor targets and PPTs.

Users can access Kimi K2.6 via the official website, app, or API. A limited-time recharge promotion (up to 30% bonus) is available for the API launch, and enterprise developers can call the official API to experience it.