whatmodel.sh / Kimi K2 Instruct (0711)
Kimi K2 Instruct (0711)
↗ share
+ contribute
Kimi K2 Instruct (0711) is an open-weights language model from Moonshot AI, released July 2025. Its published size is 1000B parameters (32B active per token) and a 131k-token context window. Licence: modified-mit.
$ how it compares
The LLMs ranked either side of Kimi K2 Instruct (0711), out of 828 with a score, then more from Moonshot AI.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 146 |  MiMo-V2-Flash (309B-A15B) | 66±1 | local | ||
| 147 |  GLM-4.6 | 66±1 | api $1.00/M +local | ||
| 148 | Kimi K2 Instruct (0711) | 66±1 | api $1.00/M +local | ||
| 149 |  o1 (2024-12-17) | 66±1 | api $26.25/M | ||
| 150 |  MiniMax-M2.5 | 66±1 | api $0.53/M +local | ||
| more from Moonshot AI | |||||
| 25 | Kimi K3 | 77±1 | api $6.00/M +local | ||
| 61 | Kimi K2.6 | 73±1 | api $1.71/M +local | ||
| 72 | Kimi K2.7 Code | 72±1 | api $1.71/M +local | ||
$ what it's good at
agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #117 of 300 · 25% of its score · 7 tests
Beats o3-pro, o3, GPT-5 mini here, though they rank higher overall. Trails SmolLM3 3B, Jamba 1.5 Large, Nova 2 Pro (Preview) here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| τ²-bench Telecom +2 | 67.5% ✎ | ██··███···██ beats 5/6 | 9.2% |
| SWE-bench Verified | 64.6% ✎ | ▁▁▁▁▁▁▂▄█▆▁▁ beats 62/131 | 5.1% |
| τ²-bench Retail +1 | 70.8% ✎ | ▅···█·▃····▃ beats 3/6 | 3.3% |
| BFCL V4 (December 2025 archive) | 59.1% ✎ | ▁▁▁▂██▅▅▆▃▂▂ beats 70/77 | 2.9% |
| τ²-bench Airline +1 | 54.2% ✎ | ▄·▄··▄···█▄▄ beats 3/6 | 2.1% |
| Terminal-Bench (pre-2.0 release) | 25.9% ✎ | ▄·▄···▄▄···█ beats 2/5 | 1.5% |
| Galileo Agent Leaderboard v2 (action completion) | 53.0% ✎ | ▅▃·▃▃▅▅█▃█▅▅ beats 15/19 | 1.2% |
coding▁▁▁▃▆██▇▆▄▃▁ #132 of 458 · 32% of its score · 5 tests
Beats Qwen3.5 Plus, Gemini 2.5 Pro, Gemini 3.1 Flash-Lite here, though they rank higher overall. Trails Devstral Small 2 24B, Llama 3.1 Nemotron Ultra 253B v1, Nemotron 3.5 Lightning 30B-A3B here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Aider polyglot (225 exercises, two attempts) | 59.1% ✎ | ▂·▂▃▅▂▃█▆▄▂▂ beats 33/49 | 16% |
| SWE-bench Verified | 64.6% ✎ | ▁▁▁▁▁▁▂▄█▆▁▁ beats 62/131 | 5.1% |
| LiveCodeBench (version unspecified) | 46.7% ✎ | █·······▄▄▄▄ beats 4/5 | 4.8% |
| MBPP+ | 81.8% ✎ | ▃·····▃█▃··▃ beats 6/6 | 3.1% |
| HumanEval+ | 86.0% ✎ | ▄▄▄·····▄▄·█ beats 0/6 | 2.8% |
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #90 of 393 * · 8.2% of its score · 2 tests
Beats LongCat 2.0 (1.6T-A48B), Inkling Small (266B-A12B), Qwen3.5 122B-A10B here, though they rank higher overall. Trails Gemma 4 12B, Kimi K2 Instruct (0905), Gemma 4 26B-A4B here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| LMArena (text) | 1418.2 ✎ | ▁▁▁▂▃▃▃▆▆▇█▃ beats 268/379 | 5.9% |
| IFEval (aggregation unspecified) | 88.9% ✎ | █··█·█··████ beats 5/6 | 2.3% |
math▁▂▃▅▆▅▄█▄▁▁▁ #133 of 398 * · 3.9% of its score · 2 tests
Beats Claude Sonnet 4.6, MiniMax-M3, MiMo-V2.5 (310B-A15B) here, though they rank higher overall. Trails Olmo 3.1 32B Think, Nemotron-Cascade-8B, Qwen3-VL 30B-A3B Thinking here, though they rank lower overall.
other tests13 not in a task · 36% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| MMLU | 89.9% ✎ | ▄·····█▄▄▄·▄ beats 3/6 | 5.9% |
| GPQA Diamond | 75.8% ✎ | █·█·█··██··█ beats 3/5 | 5.7% |
| MMLU-Pro | 82.1% ✎ | █▄·▄···▄··▄▄ beats 2/6 | 4.8% |
| EQ-Bench 3 (emotional intelligence, judge Elo) | 1311.0 ✎ | ▁·▂▁▄▅▅▃█▄▁▁ beats 52/62 | 3.6% |
| CMMLU | 89.7% ✎ | █····▄··▄▄▄▄ beats 6/6 | 3.0% |
| C-Eval | 91.3% ✎ | █····▄·▄·▄▄▄ beats 5/6 | 2.7% |
| VitaBench | 18.2% ✎ | █·······▄█·█ beats 2/6 | 2.7% |
| COLLIE (constrained text generation) | 56.3% ✎ | ▄···▄█▄····█ beats 5/6 | 2.4% |
5 more tests
| test | result | vs the others in its run | counts |
|---|---|---|---|
| DROP | 89.0% ✎ | ▄···▄·▄█·▄·▄ beats 6/6 | 2.1% |
| ZebraLogic | 89.1% ✎ | █···█··███·█ beats 3/5 | 1.0% |
| BeyondAIME | 36.6% ✎ | █····█·▄▄··▄ beats 3/6 | 0.8% |
| Open PT-LLM Leaderboard average (9 Portuguese tasks) | 81.8% ✎ | ▃▁▁▁▁▁▁▁▂█▅▃ beats 266/276 | 0.7% |
| Vectara hallucination leaderboard, 2025 dataset (HHEM-2.3, factual consistency rate) | 98.9% ✎ | ▁▁▂▃▄▄█▆▃▂▃▁ beats 137/140 | 0.2% |
$ where to run it
Through an API
api · Novita AI · $1.00/M · $0.57 in / $2.3 out per 1M tokens · 131k context · novita.ai ↗ · price sourceUSD per 1M text tokens, standard uncached input and output. Model endpoint moonshotai/kimi-k2-instruct. July 2025 Kimi K2 Instruct endpoint; not K2-0905 or K2 Thinking. Product: Serverless chat completions. API model: moonshotai/kimi-k2-instruct. Settings: service tier=standard, batch=no, cache=uncached input. Additional/excluded charges: Tools, cache storage, priority service and other modalities excluded.. Currency: USD.
llama.cpp · recommended
version
- Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.macOS / Linux
curl -LsSf https://llama.app/install.sh | shWindowsirm https://llama.app/install.ps1 | iex - Run Kimi K2 Instruct (0711) · needs ~600.5 GB · from huggingface.co/unslothchat in the terminal
llama cli -hf unsloth/Kimi-K2-Instruct-GGUF:UD-Q4_K_XL - Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf unsloth/Kimi-K2-Instruct-GGUF:UD-Q4_K_XL
LM Studio
version
- Install LM Studio · Open the app once before using the lms command.
- Run Kimi K2 Instruct (0711) · needs ~600.5 GB · from huggingface.co/unsloth
- Or download it in the terminal
lms get https://huggingface.co/unsloth/Kimi-K2-Instruct-GGUF@UD-Q4_K_XL -y - Then run lms ls to see its name, and lms chat followed by that name
MLX (Mac)
- Install MLX (Mac) · Apple silicon Macs only, text only.with uv
uv tool install mlx-lmorpip install mlx-lm - Run Kimi K2 Instruct (0711) · needs ~577.6 GB, Mac only · from huggingface.co/mlx-communitychat in the terminal
mlx_lm.chat --model mlx-community/Kimi-K2-Instruct-4bit - Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model mlx-community/Kimi-K2-Instruct-4bit
Official weights and documentation: huggingface.co/moonshotai ↗
$ how the score was worked out · Meituan LongCat-Flash-Chat mo… 63%, Aider polyglot benchmark runs 16%, Moonshot AI Kimi-K2-Instruct… 6% and 6 more
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 Meituan LongCat-Flash-Chat model card (evaluation table) | 21 | 63% |
| Aider polyglot benchmark runs | 1 | 16% |
| Moonshot AI Kimi-K2-Instruct model card (instruction-model evaluation table) · lab-reported | 1 | 6% |
 LMArena leaderboard dataset | 1 | 6% |
 EQ-Bench 3 canonical leaderboard results | 1 | 4% |
| Berkeley Function Calling Leaderboard result archive | 1 | 3% |
| Galileo Agent Leaderboard (v1 and v2 results) | 1 | 1% |
| Open Portuguese LLM Leaderboard (archived) | 1 | 1% |
| Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository) | 1 | 0% |
| prior · pull toward the middle | 0% | |
| weighted average | = 66.0 |
not counted (3)
| τ²-bench leaderboard: Moonshot AI's own submission | τ²-bench Retail [correct] | 70.6% | the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts |
| τ²-bench leaderboard: Moonshot AI's own submission | τ²-bench Airline [correct] | 56.5% | the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts |
| τ²-bench leaderboard: Moonshot AI's own submission | τ²-bench Telecom [correct] | 65.8% | the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json