# Kimi-K3

Kimi K3 is an open-weight, native multimodal agentic model and our most capable model to date. It is a 2.8T-parameter model built on Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), with native vision capabilities and a 1-million-token context window. It is the world's first open…

- **Kind:** model (Hugging Face)
- **Hugging Face:** https://huggingface.co/moonshotai/Kimi-K3
- **Owner:** moonshotai (https://forefy.com/u/hf/moonshotai)
- **Parameters:** 2779.9B
- **Context:** 1048576 tokens
- **License:** other
- **Focus:** general
- **Architecture:** KimiK3ForConditionalGeneration
- **Pinned revision:** `f831ab66814297da540d832a5235f8e904f29d06`
- **Page:** https://forefy.com/models/e0188052-b18f-5211-9592-e2289113688d
- **API (JSON):** https://forefy.com/api/asr/models/e0188052-b18f-5211-9592-e2289113688d

## Checks (repo at the pinned revision; not a behavior test)

- **Identity** [pass]: No lookalike name or conflicting provenance
- **Weights format** [pass]: safetensors only, nothing runs on load
- **Pickle files** [pass]: None.
- **Remote code** [warn]: trust_remote_code runs configuration_kimi_k3.py, modeling_kimi_k3.py on load (6 other script(s) aren't loaded automatically)
- **Chat template** [info]: None shipped
- **Refusals modified** [pass]: No
- **License** [pass]: other
- **Lineage** [info]: No base model declared
- **Model card** [warn]: 2 of 6, missing training data, language, evaluation results, intended use and limitations
- **Safetensors headers** [pass]: Every byte is indexed, no hidden data (8 of 96 files read)

## Variants

- mixed safetensors (safetensors, 1560.9 GB)

## Run it

- transformers: `AutoModelForCausalLM.from_pretrained("moonshotai/Kimi-K3", revision="f831ab66814297da540d832a5235f8e904f29d06")`
- Needs `trust_remote_code=True`, which runs the repo's Python on load. Review it first.