MO
Moonshot
Language model · China ·Nov 2025

Kimi K2 Thinking

Language Open (Restricted)
1T
Parameters
47
Benchmarks

About Kimi K2 Thinking

Kimi K2 Thinking 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 Thinking's full specifications, including 47 benchmark results. Review provider pricing and benchmark scores below, or compare Kimi K2 Thinking head-to-head with other language models.

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

Links

Announcement ↗ Hugging Face ↗

Run Kimi K2 Thinking locally — what it takes

Kimi K2 Thinking (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 Thinking 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 Thinking benchmark scores▶
Compare on the benchmark table →
Benchmark Category Score Variant
BrowseComp agentic 60.2% thinking-w-tools
BrowseComp-Zh agentic 62.3 thinking-w-tools
BrowseComp-Zh agentic 62.3 thinking-cited-deepseek-v3-2
FinSearchComp-T3 agentic 47.4 thinking-w-tools
MCP-Mark agentic 20.4 thinking-cited-deepseek-v3-2
MCP-Universe agentic 35.6 thinking-cited-deepseek-v3-2
Seal-0 agentic 56.3 thinking-w-tools
Terminal-Bench 2.0 agentic 47.1% thinking-w-simulated-tools
Terminal-Bench 2.0 agentic 35.7% thinking-cited-deepseek-v3-2
Terminal-Bench 2.0 agentic 35.7% cited-devstral-readme
Tool-Decathlon agentic 17.6 thinking-cited-deepseek-v3-2
τ²-Bench (TAU2) agentic 74.3 thinking-cited-deepseek-v3-2
LiveCodeBench code 82.6% thinking-cited-deepseek-v3-2
LiveCodeBench v6 code 83.1% thinking-no-tools
SWE-bench Verified code 71.3% thinking-cited-deepseek-v3-2
SWE-bench Verified code 71.3% cited-devstral-readme
SWE-bench Verified code 71.3% thinking-w-tools
SciCode code 44.8% thinking-no-tools
Multi-SWE-bench coding 41.9 thinking-w-tools
OJ-Bench (cpp) coding 48.7 thinking-cpp-no-tools
SWE-Bench Multilingual coding 61.1 thinking-w-tools
SWE-Bench Multilingual coding 61.1 thinking-cited-deepseek-v3-2
SWE-Bench Multilingual coding 61.1 cited-devstral-readme
Longform Writing general 73.8 thinking-no-tools
MMLU-Pro general 84.6% thinking-cited-deepseek-v3-2
MMLU-Pro general 84.6% thinking-no-tools
MMLU-Redux general 94.4% thinking-no-tools
Frames long-context 87 thinking-w-tools
AIME 2025 math 100% thinking-heavy
AIME 2025 math 99.1% thinking-w-python
AIME 2025 math 94.5% thinking-cited-deepseek-v3-2
AIME 2025 math 94.5% thinking-no-tools
HMMT February 2025 math 97.5% thinking-heavy
HMMT February 2025 math 95.1% thinking-w-python
HMMT February 2025 math 89.4% thinking-no-tools
HMMT February 2025 math 89.4% thinking-cited-deepseek-v3-2
HMMT November 2025 math 89.2 thinking-cited-deepseek-v3-2
IMOAnswerBench math 78.6 thinking-no-tools
IMOAnswerBench math 78.6 thinking-cited-deepseek-v3-2
GPQA Diamond reasoning 84.5% thinking-cited-deepseek-v3-2
GPQA Diamond reasoning 84.5% thinking-no-tools
Humanity's Last Exam reasoning 51% thinking-heavy
Humanity's Last Exam reasoning 44.9% thinking-w-tools
Humanity's Last Exam reasoning 44.9% thinking-search-agent-cited-deepseek-v3-2
Humanity's Last Exam reasoning 23.9% thinking-text-only-cited-deepseek-v3-2
Humanity's Last Exam reasoning 23.9% thinking-no-tools
HealthBench safety 58% thinking-no-tools

More models like Kimi K2 Thinking

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 Thinking? ▶

Kimi K2 Thinking 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 Thinking? ▶

Kimi K2 Thinking was developed by Moonshot and released in 2025.

Is Kimi K2 Thinking open source or proprietary? ▶

Kimi K2 Thinking 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 Thinking cost? ▶

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

How does Kimi K2 Thinking perform on benchmarks? ▶

Kimi K2 Thinking is benchmarked across 47 evaluations on GenAIList, including BrowseComp (60.2%). See the full benchmark table below and compare it with other models.

Kimi K2 Thinking è PAZZESCO... (GRANDE AGGIORNAMENTO)
Kimi K2 Thinking: BEST Opensource Model! BEATS SONNET 4.5 & GPT 5! Powerful, Fast, & Cheap!
Kimi K2 Thinking is CRAZY... (HUGE UPDATE)
NEW Top Open Model - Kimi K2 Thinking
Moonshot AI Releases Kimi K2 Thinking: A SOTA Thinking Model with Ground Breaking Performance

Know a great video about Kimi K2 Thinking?