Moonshot AI model product
Kimi K2 Thinking is the latest, most capable version of open-source thinking model. Starting with Kimi K2, it is built as a thinking agent that reasons step-by-step while dynamically invoking tools. It sets a new state-of-the-art on Humanity's Last Exam (HLE), BrowseComp, and other benchmarks by dramatically scaling multi-step reasoning depth and maintaining stable tool-use across 200–300 sequential calls. At the same time, K2 Thinking is a native INT4 quantization model with 256k context window, achieving lossless reductions in inference latency and GPU memory usage. Key features include deep thinking & tool orchestration with end-to-end training to interleave chain-of-thought reasoning with function calls, native INT4 quantization via Quantization-Aware Training (QAT) achieving lossless 2x speed-up, and stable long-horizon agency maintaining coherent goal-directed behavior across up to 200–300 consecutive tool invocations.
Updated Aug 10, 2026. Default version: Kimi K2-Thinking-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-Thinking-0905.
| FinSearchComp-T3 | 0.5 | 1 | 1 | 100.0% | C | |
| FRAMES | 0.9 | 1 | 2 | 100.0% | C | |
| HealthBench | 0.6 | 2 | 9 | 87.5% | C | |
| OJBench | 0.5 | 2 | 9 | 87.5% | C | |
| Seal-0 | 0.6 | 2 | 6 | 80.0% | C | |
| Terminal-Bench | 0.5 | 3 | 25 | 91.7% | C | |
| HMMT 2025 | 1.0 | 4 | 33 | 90.6% | C | |
| Multi-SWE-Bench | 0.4 | 4 | 6 | 40.0% | C | |
| AIME 2025 | 1.0 | 5 | 114 | 96.5% | C | |
| MMLU-Redux | 0.9 | 5 | 48 | 91.5% | C | |
| BrowseComp-zh | 0.6 | 8 | 13 | 41.7% | C | |
| SciCode | 0.4 | 8 | 18 | 58.8% | C | |
| LiveCodeBench v6 | 0.8 | 13 | 53 | 76.9% | C | |
| WritingBench | 0.7 | 15 | 15 | 0.0% | C | |
| IMO-AnswerBench | 0.8 | 17 | 19 | 11.1% | C | |
| Humanity's Last Exam | 0.5 | 18 | 92 | 81.3% | C | |
| MMLU-Pro | 0.8 | 24 | 129 | 82.0% | C | |
| SWE-bench Multilingual | 0.6 | 25 | 34 | 27.3% | C | |
| BrowseComp | 0.6 | 36 | 58 | 38.6% | C | |
| SWE-Bench Verified | 0.7 | 54 | 104 | 48.5% | C | |
| GPQA | 0.8 | 58 | 233 | 75.4% | 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.
The default version has no provider offering with current input or output token prices.
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-Thinking-0905 | 59.4 | 1T | 262.1K | 262.1K | No | MIT |
A concise description based on the published model registry.
Kimi K2 Thinking is the latest, most capable version of open-source thinking model. Starting with Kimi K2, it is built as a thinking agent that reasons step-by-step while dynamically invoking tools. It sets a new state-of-the-art on Humanity's Last Exam (HLE), BrowseComp, and other benchmarks by dramatically scaling multi-step reasoning depth and maintaining stable tool-use across 200–300 sequential calls. At the same time, K2 Thinking is a native INT4 quantization model with 256k context window, achieving lossless reductions in inference latency and GPU memory usage. Key features include deep thinking & tool orchestration with end-to-end training to interleave chain-of-thought reasoning with function calls, native INT4 quantization via Quantization-Aware Training (QAT) achieving lossless 2x speed-up, and stable long-horizon agency maintaining coherent goal-directed behavior across up to 200–300 consecutive tool invocations.
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 Thinking.
Kimi K2 Thinking's default version was released on Sep 5, 2025.
No official standard PAYG price is currently available for Kimi K2 Thinking.
Kimi K2 Thinking is published under Moonshot AI in the model registry.
The default version has a 262.1K token context window.
No. The default version is not marked as having publicly available weights.
No published provider offering is currently linked to the default version.
Nearby ranked alternatives include Qwen3.5 122B A10B, LongCat Flash Thinking, Nemotron Nano 9B.