Skip to content

EthenEthenEthen

LoginPlatformTry Ethen

Open Source Model Profile · moonshotai

Kimi-K2-Instruct

Kimi-K2-Instruct is a trillion-scale MoE text model from moonshotai. Its model card documents 32B activated parameters and chat plus agentic tool use.

Publisher
moonshotai
Task
text-generation
Model type
kimi_k2
License
modified-mit
Library
transformers
Publication status
Approved for indexing

Model overview

Kimi-K2-Instruct is published by moonshotai as a text-generation model. Safetensors metadata reports 1,026,408,235,864 parameters, with DeepseekV3ForCausalLM and model type kimi_k2. According to the model card, it is the post-trained Kimi K2 variant for chat and agentic use, described as a mixture-of-experts model.

Recorded capabilities

MoE scale description

According to the model card, Kimi K2 is a mixture-of-experts model with 1T total parameters, 32B activated parameters, 61 layers, and 128K context length.

MuonClip training account

The model card says the 1T-parameter MoE was pre-trained on 15.5T tokens and describes MuonClip optimization for stability at scale.

Agentic tool-use support

The model card documents tool-calling usage with an OpenAI-compatible client and says the inference engine must support Kimi-K2 native tool-parsing logic.

Use cases in the source record

  • General-purpose chat completion through an OpenAI-compatible chat endpoint with the publisher's default system prompt.
  • Tool-calling workflows that use the card's function-schema example on an engine with Kimi-K2 native tool-parsing support.

Limitations and unknowns

  • Provider state is historical snapshot data, not independently refreshed current availability.
  • All benchmark figures in the model card are publisher-reported comparisons and were not independently verified by Ethen.
  • No local hardware requirement is provided in the current evidence beyond the publisher's listed inference engines.

Source and provenance

Source: moonshotai/Kimi-K2-Instruct

Captured: Unknown. Processed: 2026-09-07T19:34:52.408136+00:00.

📰 Tech Blog | 📄 Paper 0. Changelog 2025.8.11 Messages with name field are now supported. We’ve also moved the chat template to a standalone file for easier viewing. 2025.7.18 We further modified our chat template to improve its robustness. The default system prompt has also been updated. 2025.7.15 We have updated our tokenizer implementation. Now special tokens like [EOS] can be encoded to their token ids. We fixed a bug in the chat template that was breaking multi-turn tool calls. 1. Model Introduction Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model with 32 billion activated parameters and 1 trillion total p…

F001F002F003F004F005F006F007F010F015F016F017F018F019F020F021F022F023F024F025F026F027F028F029F032