# MoonshotAI/Kimi-VL

Kimi-VL: Mixture-of-Experts Vision-Language Model for Multimodal Reasoning, Long-Context Understanding, and Strong Agent Capabilities

Repository: https://github.com/MoonshotAI/Kimi-VL
Canonical: https://ross.abutalabs.com/products/kimi-vl
License: MIT
License Family: permissive
Last push: 2025-07-15T15:48:21+00:00

## Health v2 (maintenance only)
Score: 33/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 31, release rhythm 35, longevity 36
- inputs: {"age_days": 511, "days_push": 414, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1224, forks 95 (observed 2026-08-28T04:04:02.821705+00:00)

## What it is
Kimi-VL is an open-source Mixture-of-Experts vision-language model (VLM) with a 2.8B activated parameter language decoder, offering multimodal reasoning, 128K long-context understanding, OCR, and agent capabilities. It includes a long-thinking variant trained with chain-of-thought SFT and reinforcement learning, with weights available on Hugging Face.

## Use cases
- run a vision-language model for image and video understanding
- extract text from images with OCR using an open model
- build multimodal agents that see screens and interact with GUIs
- reason over long documents and long videos with a 128K context window
- perform mathematical reasoning over charts and images
- compare high-resolution screenshots and UI elements
- deploy an efficient small VLM locally on GPU

## When to choose
- you need an efficient open-weight VLM with strong reasoning and long-context support
- you want multimodal OCR, document, or video comprehension without flagship-model costs
- you need agent capabilities like GUI grounding (ScreenSpot-Pro) in a compact model
- you want MIT-licensed model weights you can fine-tune or self-host

## When to avoid
- you need text-only LLM features with no vision input
- you lack GPU resources for inference even at 2.8B activated parameters
- you need a production API with SLA rather than self-managed model weights
- your task requires the very highest accuracy regardless of compute cost

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, llm-inference, ocr, computer-vision, nlp, agent-framework
- domain: large-language-models, computer-vision, artificial-intelligence, machine-learning
- platform: python, cross-platform
- tags: vision-language-model, mixture-of-experts, multimodal, long-context, chain-of-thought, model-weights, huggingface, natural-language-processing, gpu

## Member repositories
- MoonshotAI/Kimi-VL (main) score 33

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:02.821705+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:15:25.446373+00:00, confidence not recorded.
  - readme: https://github.com/MoonshotAI/Kimi-VL (fetched 2026-08-28T04:04:02.821705+00:00, sha 18711f017d8b)
- Data as of 2026-08-30T08:39:29.467469+00:00.