
Kimi K2 Instruct 0905
The September 2025 K2 refresh: 256K context, stronger agentic coding, 1T MoE open weights.
About
Released September 5, 2025, Kimi-K2-Instruct-0905 is Moonshot AI's refined follow-up to the July 2025 Kimi K2 — the 1-trillion-parameter mixture-of-experts model (32B active per token) that became the most-downloaded model on Hugging Face at launch. The 0905 update doubled context from 128K to 256K tokens and lifted performance on public benchmarks and real-world coding-agent tasks, approaching Claude Sonnet 4 levels on SWE-bench-style evals, with improved tool calling and front-end code generation. Weights remain under a modified MIT license, and it integrates well with agent scaffolds like Claude Code and Roo Code.\n\nLimitations: the modified MIT license is not a plain MIT grant — review its terms before commercial redistribution. At 1T total parameters it demands serious multi-GPU infrastructure to self-host; most developers use it via API. Moonshot iterated fast (K2 Thinking, K2.5, K2.6, K3 followed), so 0905 is a point-in-time checkpoint, not the current flagship.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Moonshot-reported: near Claude Sonnet 4 on SWE-bench-class coding evals
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 50; Trending score 88/100 from researcher assessment.