Moonshot AI releases Kimi K2.6 for long-horizon coding

Background

Kimi K2.6 is a model from Moonshot AI whose main focus is software development, released in April 2026 ^[1][2]. Its target is work that unfolds over many steps rather than a single question and answer, such as building a project across several languages or turning a prompt into a working interface ^[2].

What happened

Moonshot AI released Kimi K2.6 in April 2026 ^[1]. It is a mixture-of-experts model with 1 trillion total parameters and 32 billion active per token, using 384 experts with 8 selected and 1 shared. Its context window is 262,144 tokens ^[2]. It was released under a modified MIT license ^[2].

Key details

The model is focused on long-horizon coding, in languages such as Rust, Go and Python, and on turning prompts and visual inputs into interactive interfaces ^[2]. Long-horizon coding means work that stretches over many steps, where the model has to keep track of a project across a long sequence of actions.

Agent Swarm scaled to 300 sub-agents and 4,000 coordinated steps. The system also includes persistent background agents for scheduled tasks, and it preserves the model's thinking across turns ^[2].

Moonshot's reported scores were 80.2 on SWE-Bench Verified, 58.6 on SWE-Bench Pro, 66.7 on Terminal-Bench 2.0, 89.6 on LiveCodeBench v6 and 54.0 on Humanity's Last Exam (HLE-Full) with tools ^[2]. Terminal-Bench tests how well a model completes tasks in a command-line environment ^[2]. Together, SWE-Bench Verified, SWE-Bench Pro and Terminal-Bench 2.0 cover fixing software issues and carrying out work in a terminal, which are the core of the model's coding focus.

Impact

The release centers on coding and agent work. Because it is an open-weight model with a 262,144-token context and a parallel agent system, developers can download and inspect it, rather than only querying it through a service ^[2].