[whatmodel.sh](../../README.md) / Kimi K2 Thinking
# Kimi K2 Thinking
Moonshot AI · released 2025-11 · 1000B params (32B active) · modified-mit
**score 66.2** ±1.1 · #122 of 732 · from 13 benchmark results across 5 sources
Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only.
## What it's good at
● marks Kimi K2 Thinking among every model on that task or test, worst to best.
task vs the others rank counts tests
agent tasks ▁▁▃▆█▄●▅▄▄▂▁ #126 of 313 23% 4
coding ▁▁▁▃▅▇█▇●▄▃▁ #133 of 469* 27% 2
chat & writing ▁▁▁▃▆▄▆█▆●▇▃ #81 of 402* 12% 1
math ▁▁▃▅█▆▄●▄▂▁▁ #110 of 431* 8.7% 2
web search ██···█····●█ #2 of 5* 3.2% 1
agent tasks · #126 of 313 · 23% of its score
Beats [o3-pro](o3-pro.md), [o3](o3.md), [GPT-5 mini](gpt-5-mini.md) here, though they rank higher overall. Trails [SmolLM3 3B](smollm3-3b.md), [Jamba 1.5 Large](jamba-1-5-large.md), [Devstral 2 (2512)](devstral-2.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
Terminal-Bench 2.0 with Terminus … 35.7% ▃··▃·▃▃██●▃▅ beats 11/14 very hard · avg 6.1% 9.2% sharp · at its level
τ²-bench aggregate 74.3% ●······▄█·▄▄ beats 0/5 medium · avg 51% 5.8% typical
Terminal-Bench 2.0 35.7% ▂▂▅▅●███▆▂▃▃ beats 9/36 very hard · avg 12% 4.6% at its level · small run
BrowseComp 41.5% ▂·▃▁▂▂●▅█▂▂▄ beats 14/46 hard · avg 19% 3.2% at its level · small run
coding · #133 of 469 · 27% of its score
Beats [Qwen3.5 Plus](qwen3-5-plus.md), [Qwen3-Coder-Next (80B-A3B)](qwen3-coder-next.md), [Gemini 2.5 Pro](gemini-2-5-pro.md) here, though they rank higher overall. Trails [Llama 3.1 Nemotron Ultra 253B v1](llama-3-1-nemotron-ultra-253b.md), [Seed-OSS 36B Instruct](seed-oss-36b.md), [Solar Pro 4](solar-pro-4.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
Arena WebDev (Code Arena) 1321.9 ▁▁▃▂●█▃▆▅▅▁▂ beats 22/129 avg 981 21% sharp
WebApp1K-Duo-React pass@1 (onekq-… 70.7% ▁··▁·▁▁▁▆█●█ beats 32/45 hard · avg 29% 6.1% at its level
chat & writing · #81 of 402 · 12% of its score
Beats [Claude Haiku 5.5](claude-haiku-5-5.md), [LongCat 2.0 (1.6T-A48B)](longcat-2-0.md), [Muse Glimmer 30B](muse-glimmer-30b.md) here, though they rank higher overall. Trails [Gemma 4 12B](gemma-4-12b.md), [Gemma 4 26B-A4B](gemma-4-26b-a4b.md), [Nemotron Cascade 2 30B-A3B](nemotron-cascade-2-30b-a3b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
LMArena (text) 1430.0 ▁▁▁▂▃▃▃▆▆●█▃ beats 285/385 avg 1260 12% blunt
math · #110 of 431 · 8.7% of its score
Beats [Claude Sonnet 4.6](claude-sonnet-4-6.md), [MiniMax-M3](minimax-m3.md), [Nemotron 3 Ultra 550B-A55B](nemotron-3-ultra-550b-a55b.md) here, though they rank higher overall. Trails [Gemini 2.5 Flash](gemini-2-5-flash.md), [GLM-4.5-Air](glm-4-5-air.md), [Granite 4.2 30B](granite-4-2-30b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
OTIS mock AIME 2024–2025 83.1% ▃▃▄▃▃▃▇●█▃▂▅ beats 119/197 very hard · avg 7.2% 6.9% sharp · noisy
HMMT November 2025 89.2% ▁▁····▁▅▃●▇▁ beats 11/23 easy · avg 66% 1.8% noisy
web search · #2 of 5 · 3.2% of its score
Beats [Kimi K2.5](kimi-k2-5.md) here, though they rank higher overall.
test result vs the others in its run difficulty counts why
BrowseComp 41.5% ▂·▃▁▂▂●▅█▂▂▄ beats 14/46 hard · avg 19% 3.2% at its level · small run
other tests (4)
test result vs the others in its run difficulty counts why
GPQA Diamond 84.2% ▁▁▁▁▂▄▆▄▅●█▅ beats 143/211 hard · avg 39% 17% sharp · at its level
Chess Puzzles (Epoch) 20.0% █▁▁▁▁▂●▃▃▂▁▁ beats 78/137 very hard · avg 5.8% 5.0% sharp · noisy
ResearchRubrics 56.2% ██●··██·█··█ beats 2/6 hard · avg 33% 4.7% at its level
GAIA (no-file subset) 75.6% ▃▃····●▃···▃ beats 3/6 medium · avg 49% 2.4% at its level
source results share of the score
LMArena leaderboard dataset 2 33% ▇▇▇▇
Epoch AI 3 29% ▇▇▇
stepfun-ai/Step-3.5-Flash model card 6 22% ▇▇▇
Terminal-Bench: Benchmarking Agents on … 1 9% ▇
WebApp1K models leaderboard 1 6% ▇
prior: pull toward the middle 1% ▇
weighted average = 66.2
Licences: LMArena leaderboard dataset ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · Epoch AI ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · stepfun-ai/Step-3.5-Flash model card ([Apache-2.0](../../licenses/stepfun-ai--Step-3.5-Flash.txt)) · Terminal-Bench: Benchmarking Agents on Hard ([CC BY 4.0](../../licenses/paper-2601.11868v1-CC-BY-4.0.txt)) · WebApp1K models leaderboard ([MIT](../../licenses/webapp1k-leaderboard-MIT.txt))
## How to run it
### Through an API
api (Amazon Bedrock) · $1.07/M blended · $0.6 in / $2.5 out per 1M tokens · [docs.aws.amazon.com](https://docs.aws.amazon.com/bedrock/latest/userguide/models-supported.html)
[price source](https://github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/amazon-bedrock/models/moonshot.kimi-k2-thinking.toml): Standard base token rate from models.dev. Provider-specific tiers, long-context surcharges, caching and promotions may differ. · [context source](https://github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/moonshotai/kimi-k2-thinking.toml)
### llama.cpp (recommended)
1. Install llama.cpp. Homebrew and winget installs name the commands llama-cli and llama-server instead. [Other ways to install](https://github.com/ggml-org/llama.cpp/blob/master/docs/install.md)
```sh
# macOS / Linux
curl -LsSf https://llama.app/install.sh | sh
# Windows
irm https://llama.app/install.ps1 | iex
```
2. Run Kimi K2 Thinking (ud-q4_k_xl, needs ~600.5 GB, from [huggingface.co/unsloth](https://huggingface.co/unsloth/Kimi-K2-Thinking-GGUF/tree/main))
```sh
llama cli -hf unsloth/Kimi-K2-Thinking-GGUF:UD-Q4_K_XL
```
3. Or start a local server and open the address it prints (http://localhost:9931)
```sh
llama serve -hf unsloth/Kimi-K2-Thinking-GGUF:UD-Q4_K_XL
```
Other versions: ud-q2_k_xl (~300.5 GB) `llama cli -hf unsloth/Kimi-K2-Thinking-GGUF:UD-Q2_K_XL` · ud-iq1_s (~150.5 GB) `llama cli -hf unsloth/Kimi-K2-Thinking-GGUF:UD-IQ1_S`
### LM Studio
1. Install LM Studio. Open the app once before using the lms command. [Download it (macOS / Windows / Linux)](https://lmstudio.ai/download)
2. Run Kimi K2 Thinking (ud-q4_k_xl, needs ~600.5 GB, from [huggingface.co/unsloth](https://huggingface.co/unsloth/Kimi-K2-Thinking-GGUF/tree/main))
```sh
lms get https://huggingface.co/unsloth/Kimi-K2-Thinking-GGUF@UD-Q4_K_XL -y
```
3. Then run lms ls to see its name, and lms chat followed by that name
Other versions: ud-q2_k_xl (~300.5 GB) `lms get https://huggingface.co/unsloth/Kimi-K2-Thinking-GGUF@UD-Q2_K_XL -y` · ud-iq1_s (~150.5 GB) `lms get https://huggingface.co/unsloth/Kimi-K2-Thinking-GGUF@UD-IQ1_S -y` · 4bit (~657.6 GB, Mac only) `lms get https://huggingface.co/mlx-community/Kimi-K2-Thinking -y`
### MLX (Mac)
1. Install MLX (Mac). Apple silicon Macs only, text only.
```sh
# with uv
uv tool install mlx-lm
# or
pip install mlx-lm
```
2. Run Kimi K2 Thinking (4bit, needs ~657.6 GB, Mac only, from [huggingface.co/mlx-community](https://huggingface.co/mlx-community/Kimi-K2-Thinking/tree/035a0cdd221ae0dca6b03120e20704a251a7bc9b))
```sh
mlx_lm.chat --model mlx-community/Kimi-K2-Thinking
```
3. Or start an OpenAI-compatible server at http://localhost:8080/v1
```sh
mlx_lm.server --model mlx-community/Kimi-K2-Thinking
```
Model repository and documentation: [huggingface.co/moonshotai](https://huggingface.co/moonshotai/Kimi-K2-Thinking)
[interactive version](https://whatmodel.sh/model/kimi-k2-thinking) · [JSON](https://whatmodel.sh/api?format=json&all=1)