The AI arena is free today

Open Superagent
MoonshotAIReleased on Jul 11, 2025

Kimi K2 Base: Benchmarks, Pricing & Context Window

Kimi K2 Base is a language model from MoonshotAI, released in July 2025.

Kimi K2 base model is a state-of-the-art mixture-of-experts (MoE) language model with 32 billion activated parameters and 1 trillion total parameters. Trained on 15.5 trillion tokens with the MuonClip optimizer, this is the foundation

Kimi K2 Base benchmarks

Capability tiers

Standing within each category, adjusted for leaderboard depth.

Real tasks performance

High-confidence performance for Kimi K2 Base across real-world prompt categories. Only 95% intervals at most 4 points wide are shown.

Performance by conversation depth

How Kimi K2 Base holds up as conversations get longer.

Quality Tracker

Kimi K2 Base Performance Across Datasets

Scores sourced from the model's scorecard, paper, or official blog posts

LLM Stats Logollm-stats.com - Wed Sep 09 2026
Notice missing or incorrect data?

Kimi K2 Base model size

Kimi K2 Base has 1 trillion parameters and was trained on 15.5 trillion tokens. See how it compares to other models in the same parameter range.

ParametersTraining tokens
1T
15.5Ttokens
16× tokens-to-params ratio
Frontier (200B+)
1T
1B7B70B405B

Try now

huggle
Kimi K2 Basein Huggle

Make it with
Kimi K2 Base.

Kimi K2 Base

Kimi K2 Base latency

Kimi K2 Base time to first token, sustained output throughput, and failed-request rate from live model usage over the trailing 7 days.

Kimi K2 Base examples

Recent arena outputs from Kimi K2 Base, picked from the highest-ranked matchups.

Kimi K2 Base license

Kimi K2 Base is released under the MIT license, which permits commercial use, has 1.0T parameters.

License
MIT
Commercial use allowed
Parameters
1.0T

MIT License - allows commercial use

Kimi K2 Base resources

Official sources for Kimi K2 Base: provider documentation, official launch post, source repository, model weights.

Kimi K2 Base vs other models

The most-compared alternatives to Kimi K2 Base are GPT-4 Turbo, Mistral Small 3.2 24B Instruct, Qwen3 VL 30B A3B Thinking. Open any pair side-by-side for benchmarks, pricing, context, and latency.

Models like Kimi K2 Base

Models ranked just above and below Kimi K2 Base by LLM Stats score.

 

GPT-4 Turbo

Score pending
 

Mistral Small 3.2 24B Instruct

Score pending
 

Qwen3 VL 30B A3B Thinking

Score pending
 

Qwen2.5 32B Instruct

Score pending
 

DeepSeek-V3

Score pending
 

Phi 4

Score pending

FAQ

Common questions about Kimi K2 Base.

When was Kimi K2 Base released?

Kimi K2 Base was released on July 11, 2025 by MoonshotAI. This is the official Kimi K2 Base release date tracked on LLM Stats.

How big is Kimi K2 Base?

Kimi K2 Base has 1000 billion parameters. It was trained on 15.5 trillion tokens. It ships as an open-weight model, so you can download and run it on your own hardware.

Who created Kimi K2 Base?

Kimi K2 Base was created by MoonshotAI.

What is the license for Kimi K2 Base?

Kimi K2 Base is released under the MIT license. This is an open-source / open-weight license that permits self-hosting.

Where is the Kimi K2 Base paper or technical report?

Kimi K2 Base has a paper or technical report available at https://moonshotai.github.io/Kimi-K2/. Use that source for architecture, training, release and evaluation details.

What models should I compare Kimi K2 Base against?

Common Kimi K2 Base comparisons include Kimi K2 Base vs GPT-4 Turbo, Kimi K2 Base vs Mistral Small 3.2 24B Instruct, Kimi K2 Base vs Qwen3 VL 30B A3B Thinking. Compare them side by side for benchmark scores, pricing, context window, latency and provider availability.