whatmodel.sh / Kimi K2 Thinking

Kimi K2 Thinking

+ contribute

Kimi K2 Thinking is an open-weights language model from Moonshot AI, released November 2025. Its published size is 1000B parameters (32B active per token) and a 262k-token context window. Licence: modified-mit.

67.0±1.1#112 of 719 LLMsfrom 13 benchmark results across 5 sources

$ how it compares

The LLMs ranked either side of Kimi K2 Thinking, out of 828 with a score, then more from Moonshot AI.

#	model	score	where	RAM	released
127	K-EXAONE 2.0 (750B-A37B) LG AI Research	67±1	local	~791GB	2026-07
128	Claude Opus 4.1 Anthropic	67±1	comparison only		2025-08
129	Kimi K2 Thinking Moonshot AI	67±1	api $1.07/M +local	~160GB	2025-11
130	Laguna M.1 Poolside	67±2	api price unknown +local	~135GB	2026-04
131	Qwen3.7 Flash Alibaba	67±1	api $0.055/M		2026-07
	more from Moonshot AI
25	Kimi K3 Moonshot AI	77±1	api $6.00/M +local	~884GB	2026-07
61	Kimi K2.6 Moonshot AI	73±1	api $1.71/M +local	~317GB	2026-04
72	Kimi K2.7 Code Moonshot AI	72±1	api $1.71/M +local	~359GB	2026-06

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #115 of 300 · 26% of its score · 4 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	difficulty	counts	why
Terminal-Bench 2.0 with Terminus 2 agent (benchmark authors' runs)	35.7% ✎	▃··▃·▃▃██▅▃▅ beats 11/14	very hard · avg 5.8%	▇▇▇▇ 10%	sharp · at its level
τ²-bench aggregate	74.3% ✎	▄······▄█·▄▄ beats 0/5	medium · avg 46%	▇▇▇ 7.3%	typical
Terminal-Bench 2.0	35.7% ✎	▂▂▅▅████▆▂▃▃ beats 9/36	very hard · avg 11%	▇▇ 5.2%	at its level · small run
BrowseComp	41.5% ✎	▂·▄▁▂▃▆▄█▂▃▄ beats 14/42	hard · avg 17%	▇▇ 3.9%	at its level · small run

coding▁▁▁▃▆██▇▆▄▃▁ #124 of 458 * · 24% of its score · 2 tests

Beats Qwen3.5 Plus, Gemini 2.5 Pro here, though they rank higher overall. Trails Llama 3.1 Nemotron Ultra 253B v1, Seed-OSS 36B Instruct, Gemma 4 26B-A4B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Arena WebDev (Code Arena)	1321.4 ✎	▁▂▄▃▅█▄▇▆▄▁▁ beats 22/119	avg 981	▇▇▇▇▇▇▇▇ 19%	sharp
WebApp1K-Duo-React pass@1 (onekq-ai)	70.7% ✎	▁··▁·▁▁▁▆█▇█ beats 32/45	hard · avg 32%	▇▇ 4.8%	at its level

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #77 of 393 * · 11% of its score · 1 test

Beats LongCat 2.0 (1.6T-A48B), Mistral Large 4 (public preview), Inkling Small (266B-A12B) here, though they rank higher overall. Trails Gemma 4 12B, Gemma 4 26B-A4B, Nemotron Cascade 2 30B-A3B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LMArena (text)	1430.1 ✎	▁▁▁▂▃▃▃▆▆▇█▃ beats 284/379	avg 1268	▇▇▇▇ 11%	blunt

math▁▂▃▅▆▅▄█▄▁▁▁ #101 of 398 * · 10% of its score · 2 tests

Beats Claude Sonnet 4.6, MiniMax-M3, Nemotron 3 Ultra 550B-A55B here, though they rank higher overall. Trails Gemini 2.5 Flash, GLM-4.5-Air, Granite 4.2 30B 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 119/193	very hard · avg 8.5%	▇▇▇ 7.8%	sharp · noisy
HMMT November 2025	89.2% ✎	▁▁····▁▅▃█▇▁ beats 11/23	medium · avg 65%	▇ 2.0%	noisy

web search██···█····██ #2 of 5 * · 3.9% of its score · 1 test

Beats Kimi K2.5 here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
BrowseComp	41.5% ✎	▂·▄▁▂▃▆▄█▂▃▄ beats 14/42	hard · avg 17%	▇▇ 3.9%	at its level · small run

other tests4 not in a task · 28% of its score

test	result	vs the others in its run	difficulty	counts	why
GPQA Diamond	84.2% ✎	▁▁▁▁▂▄▆▄▅▇█▅ beats 143/208	hard · avg 40%	▇▇▇▇▇▇▇ 16%	sharp · at its level
ResearchRubrics	56.2% ✎	███··██·█··█ beats 2/6	hard · avg 31%	▇▇ 5.1%	at its level
Chess Puzzles (Epoch)	20.0% ✎	█▁▁▁▁▂▁▄▂▂▁▁ beats 78/135	very hard · avg 5.9%	▇▇ 4.3%	sharp · noisy
GAIA (no-file subset)	75.6% ✎	▃▃····█▃···▃ beats 3/6	medium · avg 47%	▇ 2.6%	at its level

$ where to run it

Through an API

api · Amazon Bedrock · $1.07/M · $0.6 in / $2.5 out per 1M tokens · 262k context · docs.aws.amazon.com ↗ · price sourceStandard base token rate from models.dev. Provider-specific tiers, long-context surcharges, caching and promotions may differ.

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 | sh
Windows
irm https://llama.app/install.ps1 | iex
other ways to install ↗
Run Kimi K2 Thinking · needs ~600.5 GB · from huggingface.co/unsloth
chat in the terminal
llama cli -hf unsloth/Kimi-K2-Thinking-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-Thinking-GGUF:UD-Q4_K_XL

LM Studio

version

Install LM Studio · Open the app once before using the lms command.
download it (macOS / Windows / Linux) ↗
Run Kimi K2 Thinking · needs ~600.5 GB · from huggingface.co/unsloth
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/unsloth/Kimi-K2-Thinking-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-lm
or
pip install mlx-lm
Run Kimi K2 Thinking · needs ~657.6 GB, Mac only · from huggingface.co/mlx-community
chat in the terminal
mlx_lm.chat --model mlx-community/Kimi-K2-Thinking
Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model mlx-community/Kimi-K2-Thinking

Official weights and documentation: huggingface.co/moonshotai ↗

$ how the score was worked out · LMArena leaderboard dataset 30%, Epoch AI — internal evaluatio… 28%, stepfun-ai/Step-3.5-Flash mod… 26% and 2 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	licence
LMArena leaderboard dataset	2	▇▇▇▇▇▇▇▇▇ 30%	CC BY 4.0
Epoch AI — internal evaluations	3	▇▇▇▇▇▇▇▇▇ 28%	CC BY 4.0
stepfun-ai/Step-3.5-Flash model card	6	▇▇▇▇▇▇▇▇ 26%	Apache-2.0
Terminal-Bench: Benchmarking Agents on Hard, Realistic Tasks in Command Line Interfaces — benchmark authors' own agent-model trials	1	▇▇▇ 10%	CC BY 4.0
WebApp1K models leaderboard (onekq-ai)	1	▇ 5%	MIT
prior · pull toward the middle		▇ 1%
weighted average		= 67.0