Moonshot AI model product
Kimi K2-Instruct-0905 is the latest, most capable version of Kimi K2, achieving state-of-the-art performance in frontier knowledge, math, and coding among non-thinking models. This Mixture-of-Experts model features 32 billion activated parameters and 1 trillion total parameters, meticulously optimized for agentic tasks. Key features include enhanced agentic coding intelligence, extended context length to 256K tokens, and a hybrid architecture trained with MuonClip optimizer on 15.5T tokens. The model achieves 65.8% on SWE-bench Verified (single attempt), 47.3% on SWE-bench Multilingual, and excels at tool use with 70.6% on Tau2-retail. It is a reflex-grade model without long thinking, designed to act and execute complex tasks seamlessly.
Updated Aug 10, 2026. Default version: Kimi K2-Instruct-0905
Structured fields from the published default version.
This profile uses the latest version under this unique model that has a calculated LLMBoard score. Arena and price are excluded.
Published benchmark records for the scored version Kimi K2-Instruct-0905.
| ACEBench | 0.8 | 2 | 2 | 0.0% | C | |
| AutoLogi | 0.9 | 2 | 2 | 0.0% | C | |
| CNMO 2024 | 0.7 | 2 | 3 | 50.0% | C | |
| PolyMath-en | 0.7 | 2 | 2 | 0.0% | C | |
| MultiPL-E | 0.9 | 5 | 13 | 66.7% | C | |
| ZebraLogic | 0.9 | 6 | 8 | 28.6% | C | |
| MATH-500 | 1.0 | 7 | 32 | 80.7% | C | |
| OJBench | 0.3 | 9 | 9 | 0.0% | C | |
| LiveBench | 0.8 | 10 | 38 | 75.7% | C | |
| Aider-Polyglot | 0.6 | 13 | 22 | 42.9% | C | |
| MMLU | 0.9 | 14 | 100 | 86.9% | C | |
| Multi-Challenge | 0.5 | 15 | 29 | 50.0% | C | |
| IFEval | 0.9 | 17 | 65 | 75.0% | C | |
| MMLU-Redux | 0.9 | 17 | 48 | 66.0% | C | |
| Tau2 Airline | 0.6 | 17 | 23 | 27.3% | C | |
| Tau2 Retail | 0.7 | 21 | 26 | 20.0% | C | |
| SuperGPQA | 0.6 | 22 | 34 | 36.4% | C | |
| Terminal-Bench | 0.3 | 23 | 25 | 8.3% | C | |
| SimpleQA | 0.3 | 25 | 46 | 46.7% | C | |
| Tau2 Telecom | 0.7 | 28 | 35 | 20.6% | C | |
| HMMT 2025 | 0.4 | 30 | 33 | 9.4% | C | |
| SWE-bench Multilingual | 0.5 | 31 | 34 | 9.1% | C | |
| LiveCodeBench | 0.5 | 40 | 73 | 45.8% | C | |
| AIME 2024 | 0.7 | 42 | 53 | 21.1% | C | |
| MMLU-Pro | 0.8 | 46 | 129 | 64.8% | C | |
| SWE-Bench Verified | 0.7 | 72 | 104 | 31.1% | C | |
| Humanity's Last Exam | 0.0 | 91 | 92 | 1.1% | C | |
| AIME 2025 | 0.5 | 103 | 114 | 9.7% | C | |
| GPQA | 0.8 | 105 | 233 | 55.2% | C |
Preference and agent-evaluation signals from published Arena datasets.
The default version has no published Arena rows, or its source alias has not been resolved.
Official vendor API PAYG pricing is summarized first. The table then lists individual provider offerings without treating their minimum as the official price.
| Hugging Face | moonshotai/Kimi-K2-Instruct-0905 | global | $1 | $3 | 262.1K |
Official prices use only the vendor's configured official Provider and positive standard USD PAYG rates. Third-party offers remain explicitly labeled.
All published versions linked to this unique model. The score columns identify the version used by the current overall ranking.
| Kimi K2 0905 | N/A | 1T | 262.1K | 262.1K | No | Proprietary | |
| Kimi K2-Instruct-0905 | 51.4 | 1T | N/A | N/A | No | MIT | |
| Kimi K2 Base | N/A | 1T | N/A | N/A | No | MIT | |
| Kimi K2 Instruct | N/A | 1T | 200K | 200K | No | MIT |
A concise description based on the published model registry.
Kimi K2-Instruct-0905 is the latest, most capable version of Kimi K2, achieving state-of-the-art performance in frontier knowledge, math, and coding among non-thinking models. This Mixture-of-Experts model features 32 billion activated parameters and 1 trillion total parameters, meticulously optimized for agentic tasks. Key features include enhanced agentic coding intelligence, extended context length to 256K tokens, and a hybrid architecture trained with MuonClip optimizer on 15.5T tokens. The model achieves 65.8% on SWE-bench Verified (single attempt), 47.3% on SWE-bench Multilingual, and excels at tool use with 70.6% on Tau2-retail. It is a reflex-grade model without long thinking, designed to act and execute complex tasks seamlessly.
Use the benchmark, Arena and pricing sections above as separate evidence. A missing field means the current data snapshot does not support that claim.
Data snapshot: 2026-08-07. Editorial model content is not available in the backend.
Open a comparison with the three ranked models immediately above and below this model.
Recommendations prioritize the same model type and family, then the closest published LLMBoard score.
Common questions about Kimi K2.
Kimi K2's default version was released on Sep 5, 2025.
No official standard PAYG price is currently available for Kimi K2. The lowest tracked third-party offer starts at $1 input and $3 output via Hugging Face.
Kimi K2 is published under Moonshot AI in the model registry.
The current registry does not publish a context window for the default version.
No. The default version is not marked as having publicly available weights.
1 published provider offerings are linked to the default version.
Nearby ranked alternatives include Llama 3.1 70B, Qwen3 VL 32B Thinking, Qwen2.5 VL 32B.