Model Details
Kimi K2 Instruct 0905
Quick answer: Kimi K2 Instruct 0905 is Moonshot AI's September 2025 checkpoint of Kimi K2 Instruct, scoring 76.5% ACEBench and 81.1% MMLU-Pro. Apache 2.0 licensed for full commercial use.
At a Glance
Where Kimi K2 Instruct 0905 leads
- 76.5% ACEBench — strong coding task performance
- 81.1% MMLU-Pro — competitive academic knowledge
- Apache 2.0 — fully open for commercial use and fine-tuning
- Updated September 2025 checkpoint over base K2 Instruct
Where it lags
- Limited public benchmark coverage beyond ACEBench and MMLU-Pro
- Proprietary training data details
- Text-only modality
Best for: Open-source coding and reasoning pipelines; teams requiring Apache 2.0 MoE models with strong coding capability.
What Kimi K2 Instruct 0905 Is
Kimi K2 Instruct 0905 is the September 2025 checkpoint of Moonshot AI's Kimi K2 Instruct model. The "0905" suffix (September 5, 2025) indicates a post-release update from the base Kimi K2 Instruct. Kimi K2 is a mixture-of-experts (MoE) model built for agentic tasks and coding, with the Instruct variant optimised for instruction-following.
With 76.5% ACEBench and 81.1% MMLU-Pro under Apache 2.0, this checkpoint is competitive for open-source deployment in coding-heavy pipelines.
Specifications
| Field | Value |
|---|---|
| Organization | Moonshot AI |
| License | Apache 2.0 |
| Release date | July 2025 (checkpoint: Sept 2025) |
| Architecture | MoE (Mixture of Experts) |
| Modality | Text only |
Pricing
Open weights under Apache 2.0 — self-host at no license cost. Also available via Moonshot AI API.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| ACEBench | 76.5% | Benchgen evaluation | 2025-09 |
| MMLU-Pro | 81.1% | Benchgen evaluation | 2025-09 |
Kimi K2 Instruct 0905 vs Alternatives
| Model | MMLU-Pro | License | Architecture |
|---|---|---|---|
| Kimi K2 Instruct 0905 | 81.1% | Apache 2.0 | MoE |
| Kimi K2 Thinking 0905 | 84.6% | Apache 2.0 | MoE (Thinking) |
| QwQ-32B | — | Apache 2.0 | Dense |
| DeepSeek-V3 | — | MIT | MoE |
Kimi K2 Instruct 0905 vs Kimi K2 Thinking 0905: lower MMLU-Pro (81.1% vs 84.6%) but likely faster inference without thinking overhead. For direct instruction-following, Instruct 0905; for advanced reasoning, Thinking 0905.
Related Models & Benchmarks
- Compare: Kimi K2 Thinking 0905 · Kimi K2 Instruct · DeepSeek-V3
- Benchmarks: ACEBench · MMLU-Pro
Frequently Asked Questions
What is Kimi K2 Instruct 0905?
Moonshot AI's September 2025 checkpoint of Kimi K2 Instruct, scoring 76.5% ACEBench and 81.1% MMLU-Pro. Apache 2.0 MoE model for coding and reasoning.What is the difference between Kimi K2 Instruct and Kimi K2 Instruct 0905?
Kimi K2 Instruct 0905 is a later checkpoint (September 5, 2025) with updated training compared to the base Kimi K2 Instruct release. The 0905 checkpoint may include improved alignment or performance updates.Specs from Moonshot AI's Kimi K2 Instruct 0905 release and Benchgen evaluations. Last updated 2026-07-24.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.