What are the hardware requirements for Kimi K2 Thinking?

Kimi K2 Thinking requires approximately 383.2 GB of VRAM and supports a context window of 256K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Kimi K2 Thinking and how large is it?

Kimi K2 Thinking is developed by moonshotai, a model with 1058.1b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does Kimi K2 Thinking perform on standard benchmarks?

Kimi K2 Thinking has the following published scores: LMSYS ELO 1430. Compare against reference models on this page or on the LLM Explorer leaderboards.

Kimi K2 Thinking by moonshotai — VRAM 383.2GB, 256K context

Name: Kimi K2 Thinking
Author: moonshotai

Kimi K2 Thinking is an open-source language model by moonshotai. Features: 1058.1b LLM, VRAM: 383.2GB, Context: 256K, License: other, LLM Explorer Score: 0.51, ELO: 1430.

Compressed-tensors Conversational Custom code Endpoints compatible Eval-results Kimi k2 Region:us Safetensors Sharded Tensorflow

Model Card on HF 🤗: https://huggingface.co/moonshotai/Kimi-K2-Thinking

Kimi K2 Thinking Benchmarks

LMSys ELO 1430 vs 1272 (so35) 12.4%

^nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").

What is the LLM Explorer Rank (Score)

Kimi K2 Thinking Parameters and Internals

LLM Name	Kimi K2 Thinking
Repository 🤗	https://huggingface.co/moonshotai/Kimi-K2-Thinking
Model Size	1058.1b
Required VRAM	383.2 GB
Updated	2026-07-08
Maintainer	moonshotai
Model Type	kimi_k2
Model Files	1.0 GB: 1-of-62 9.8 GB: 2-of-62 9.8 GB: 3-of-62 9.8 GB: 4-of-62 9.8 GB: 5-of-62 9.8 GB: 6-of-62 9.8 GB: 7-of-62 9.8 GB: 8-of-62 9.8 GB: 9-of-62 9.8 GB: 10-of-62 9.8 GB: 11-of-62 9.8 GB: 12-of-62 9.8 GB: 13-of-62 9.8 GB: 14-of-62 9.8 GB: 15-of-62 9.8 GB: 16-of-62 9.8 GB: 17-of-62 9.8 GB: 18-of-62 9.8 GB: 19-of-62 9.8 GB: 20-of-62 9.8 GB: 21-of-62 9.8 GB: 22-of-62 9.8 GB: 23-of-62 9.8 GB: 24-of-62 9.8 GB: 25-of-62 9.8 GB: 26-of-62 9.8 GB: 27-of-62 9.8 GB: 28-of-62 9.8 GB: 29-of-62 9.8 GB: 30-of-62 9.8 GB: 31-of-62 9.8 GB: 32-of-62 9.8 GB: 33-of-62 9.8 GB: 34-of-62 9.8 GB: 35-of-62 9.8 GB: 36-of-62 9.8 GB: 37-of-62 9.8 GB: 38-of-62 9.8 GB: 39-of-62 9.8 GB: 40-of-62
Model Architecture	DeepseekV3ForCausalLM
License	other
Context Length	262144
Model Max Length	262144
Transformers Version	4.51.3
Vocabulary Size	163840
Torch Data Type	bfloat16

Rank the Kimi K2 Thinking capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Quantized Models of the Kimi K2 Thinking

Model	Likes	Downloads	VRAM
Kimi K2 Thinking 4bit	9	589	151 GB
Kimi K2 Thinking 3bit	1	140	219 GB

Best Alternatives to Kimi K2 Thinking

Best Alternatives	Context / RAM	Downloads	Likes
Kimi K2 Thinking	256K / 383.2 GB	77	4