MO
Moonshot
Language model · China ·Jul 2025

Kimi K2

Language Open (Restricted)
1T
Parameters
68
Benchmarks

About Kimi K2

Kimi K2 is an AI model developed by Moonshot, in the language category, released in 2025, made available as an open-weights model with 1000B params.

On this page you'll find Kimi K2's full specifications, including 68 benchmark results. Review provider pricing and benchmark scores below, or compare Kimi K2 head-to-head with other language models.

→ More models by Moonshot → All language models → AI benchmarks leaderboard

Links

Announcement ↗ Hugging Face ↗

Run Kimi K2 locally — what it takes

Kimi K2 (1,000B) is too large to run on a single consumer or Apple machine. At Q4 its weights alone are about 560 GB — it needs a multi-GPU / datacenter setup:

  • ≈ 3× AMD Instinct MI325X
  • ≈ 4× NVIDIA B200

Lower quantization (Q3/Q2) or CPU + system-RAM offload (e.g. ktransformers) reduces the requirement at the cost of speed.

Software support

·ExLlamaV2 ·KoboldCpp ·LM Studio ·MLX ·Ollama ·SGLang ·TensorRT-LLM ·llama.cpp ·vLLM ?KTransformers ?text-generation-webui

✓ measured · · compatible · — not supported. Informational only — speed is hardware-based.

Community

Would you run Kimi K2 again?
No votes yet — be the first.
Did it run? — community reports by hardware

No reports yet. Own this model on some hardware? Be the first to confirm it runs.

Discussion

No comments yet. Start the discussion.

Kimi K2 benchmark scores▶
Compare on the benchmark table →
Benchmark Category Score Variant
BrowseComp agentic 14.1% react-cited-tongyi
BrowseComp agentic 7.9%
BrowseComp agentic 7.4% k2-0905-w-tools-cited-k2-thinking
BrowseComp-Zh agentic 28.8 react-cited-tongyi
BrowseComp-Zh agentic 22.2 k2-0905-w-tools-cited-k2-thinking
FinSearchComp-T3 agentic 10.4 k2-0905-w-tools-cited-k2-thinking
GAIA agentic 57.7 react-cited-tongyi
Seal-0 agentic 25.2 k2-0905-w-tools-cited-k2-thinking
TAU-bench Airline agentic 51.2%
TAU-bench Retail agentic 73.9% reported in GLM-4.5 paper
Terminal-Bench 2.0 agentic 44.5% k2-0905-w-simulated-tools-cited-k2-thinking
Terminal-Bench 2.0 agentic 25%
WebWalkerQA agentic 63 react-cited-tongyi
xbench-DeepSearch agentic 50 react-cited-tongyi
Codeforces Elo code 1,574 instruct-0905-rating-cited-ling-1t
FullStackBench code 54% instruct-0905-cited-ling-1t
HumanEval code 87.1% pass@1
LiveCodeBench code 48.95% instruct-0905-cited-ling-1t
LiveCodeBench v6 code 56.1% k2-0905-no-tools-cited-k2-thinking
MBPP code 89.96% instruct-0905-sanitized-cited-ling-1t
SWE-bench Verified code 69.2% k2-0905-w-tools-cited-k2-thinking
SWE-bench Verified code 65.8%
SWE-bench Verified code 65.4%
SciCode code 30.7% k2-0905-no-tools-cited-k2-thinking
Aider Benchmark coding 85.34 instruct-0905-edit-cited-ling-1t
Multi-SWE-bench coding 33.5 k2-0905-w-tools-cited-k2-thinking
MultiPL-E coding 73.54 instruct-0905-cited-ling-1t
OJ-Bench (cpp) coding 25.5 k2-0905-cpp-no-tools-cited-k2-thinking
SWE-Bench Multilingual coding 55.9 k2-0905-w-tools-cited-k2-thinking
ArenaHard v2 general 69.88 instruct-0905-win-rate-cited-ling-1t
C-Eval general 91.12% instruct-0905-cited-ling-1t
IFEval general 90.99% instruct-0905-prompt-strict-cited-ling-1t
IFEval general 89.8%
Longform Writing general 62.8 k2-0905-no-tools-cited-k2-thinking
MMLU general 89.5%
MMLU-Pro general 81.9% k2-0905-no-tools-cited-k2-thinking
MMLU-Pro general 81.03% instruct-0905-cited-ling-1t
MMLU-Redux general 92.7% k2-0905-no-tools-cited-k2-thinking
MMLU-Redux general 91.58% instruct-0905-cited-ling-1t
MultiChallenge general 54.1%
MultiChallenge general 48.72% instruct-0905-cited-ling-1t
Frames long-context 72 react-cited-tongyi
Frames long-context 58.1 k2-0905-w-tools-cited-k2-thinking
AIME 2024 math 67.24% instruct-0905-cited-ling-1t
AIME 2025 math 75.2% k2-0905-w-python-cited-k2-thinking
AIME 2025 math 51% k2-0905-no-tools-cited-k2-thinking
AIME 2025 math 50.16% instruct-0905-cited-ling-1t
CNMO 2024 math 68.92% instruct-0905-cited-ling-1t
HMMT February 2025 math 70.4% k2-0905-w-python-cited-k2-thinking
HMMT February 2025 math 38.8% instruct-0905-cited-ling-1t
HMMT February 2025 math 38.8% k2-0905-no-tools-cited-k2-thinking
IMOAnswerBench math 45.8 k2-0905-no-tools-cited-k2-thinking
Omni-MATH math 62.42 instruct-0905-cited-ling-1t
UGMathBench math 69.97 instruct-0905-cited-ling-1t
ARC-AGI-1 reasoning 22.19 instruct-0905-cited-ling-1t
BBEH reasoning 34.83 instruct-0905-cited-ling-1t
FinanceReasoning reasoning 84.83 instruct-0905-cited-ling-1t
GPQA Diamond reasoning 74.2% k2-0905-no-tools-cited-k2-thinking
GPQA Diamond reasoning 73.93% instruct-0905-cited-ling-1t
Humanity's Last Exam reasoning 21.7% k2-0905-w-tools-cited-k2-thinking
Humanity's Last Exam reasoning 18.1% react-cited-tongyi
Humanity's Last Exam reasoning 7.9% k2-0905-no-tools-cited-k2-thinking
Humanity's Last Exam reasoning 7.29% instruct-0905-cited-ling-1t
ZebraLogic reasoning 85.5% instruct-0905-cited-ling-1t
HealthBench safety 43.8% k2-0905-no-tools-cited-k2-thinking
SimpleQA safety 31%
BFCL v3 tool-use 71.1%
BFCL v3 tool-use 71.05% instruct-0905-cited-ling-1t

More models like Kimi K2

AI
DMind-1
 AI
COLLECTIVE-1
 AI
Cogito 1 series
 AI
Euler (Euler大模型)
· 2.5B
 AI
SmolLM-1.7B
· 1.7B
 AI
Llama-3-70B-Synthia-v3.5
· 70B

Frequently asked questions

What is Kimi K2? ▶

Kimi K2 is an AI model developed by Moonshot, in the language category, released in 2025. It is tracked on GenAIList with its specifications, benchmark scores and provider pricing.

Who created Kimi K2? ▶

Kimi K2 was developed by Moonshot and released in 2025.

Is Kimi K2 open source or proprietary? ▶

Kimi K2 ships as an open-weights model: you can download and self-host the weights, though the licence may place some restrictions on use.

How much does Kimi K2 cost? ▶

Pricing for Kimi K2 depends on the provider. See the providers table on this page for the latest API rates.

How does Kimi K2 perform on benchmarks? ▶

Kimi K2 is benchmarked across 68 evaluations on GenAIList, including BrowseComp (14.1%). See the full benchmark table below and compare it with other models.

Kimi K2 explained in 5 minutes
Kimi K2 is INSANE... (Open-Source is BACK!)
The World's BEST New AI Model is 100% Free (Kimi K2 Thinking)
Kimi K2 AI Model | Advanced Open-Source Performance | Moonshot AI
Kimi K2 Reasoning (Fully Tested): IS IT REALLY THE BEST Open Model?

Know a great video about Kimi K2?