selfhost.directory

The index of self-hostable projects — ranked, categorized and served with live updates.
Own your data and stop giving it to big tech.

Alternatives to Hy4-flash-base

120 open-weight models that do the same job as Hy4-flash-base (219B vision), ranked by how close they are in task and size, then by popularity and how recent they are. Licence and VRAM are shown for each.

the original
tencent logo

Hy4-flash-base

tencent·Vision

≈133 GB

Hy4 flash is our most cost-effective Hy model to date, featuring 220B total parameters (including 64B Engram parameters) with just 12B active parameters per token. It natively accepts text and image inputs, generates text, and supports a context window of up to 1M tokens.

219B1M ctx tencent-hy-license
699Nov 2026
--vram
any
--gpus
--quant
  1. Qwen logo

    Qwen3.5-9B

    Qwen·Vision

    ≈7.2 GB

    Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance.

    9.7B256K ctx Apache 2.0
    6.3MFeb 2026ollama
    • Same category
    • Smaller (9.7B vs 219B)
    • Commercial use OK
    • Widely used
    vs
  2. google logo≈18 GB

    DiffusionGemma is a generative model built by Google DeepMind. Based on the 26B A4B Mixture-of-Experts (MoE) Gemma 4 architecture, DiffusionGemma generates tokens using discrete diffusion. This open-weights model is multimodal, handling text, image, and video inputs to generate text output.

    25.8BA4B256K ctx Apache 2.0
    581KJun 2026gguf
    • Same category
    • Smaller (25.8B vs 219B)
    • Commercial use OK
    vs
  3. Qwen logo

    Qwen3.5-2B

    Qwen·Vision

    ≈2.2 GB

    Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance.

    2.3B256K ctx Apache 2.0
    5.5MFeb 2026ollama
    • Same category
    • Smaller (2.3B vs 219B)
    • Commercial use OK
    • Widely used
    vs
  4. huihui-ai logo≈19 GB

    This is an uncensored version of Qwen/Qwen3.8-27B created with abliteration (see remove-refusals-with-transformers to know more about it). This is a crude, proof-of-concept implementation to remove refusals from an LLM model without using TransformerLens.

    27.8B256K ctx Apache 2.0
    160KAug 2026gguf
    • Same category
    • Smaller (27.8B vs 219B)
    • Commercial use OK
    vs
  5. baidu logo

    Qianfan-VL-70B

    baidu·Vision

    ≈46 GB

    The models in this series, including the 4B-parameter end-to-end vision-language model, are presented in the paper Qianfan-OCR: A Unified End-to-End Model for Document Intelligence.

    71.8B32K ctx MIT
    320Sept 2025
    • Same category
    • Smaller (71.8B vs 219B)
    • Commercial use OK
    vs
  6. Qwen logo

    Qwen3.5-4B

    Qwen·Vision

    ≈4.3 GB

    Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance.

    4.7B256K ctx Apache 2.0
    4.4MFeb 2026ollama
    • Same category
    • Smaller (4.7B vs 219B)
    • Commercial use OK
    • Widely used
    vs
  7. Qwen logo≈22 GB

    Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance.

    36BA3B256K ctx Apache 2.0
    42.5KFeb 2026gguf
    • Same category
    • Smaller (36B vs 219B)
    • Commercial use OK
    vs
  8. Qwen logo

    QVQ-72B-Preview

    Qwen·Vision

    ≈47 GB

    QVQ-72B-Preview is an experimental research model developed by the Qwen team, focusing on enhancing visual reasoning capabilities.

    73.4B125K ctx Qwen License
    1.6KDec 2024gguf
    • Same category
    • Smaller (73.4B vs 219B)
    vs
  9. Qwen logo≈22 GB

    Meet Qwen3-VL — the most powerful vision-language model in the Qwen series to date.

    33.4B256K ctx Apache 2.0
    217KOct 2025ollama
    • Same category
    • Smaller (33.4B vs 219B)
    • Commercial use OK
    vs
  10. Cloudflare logo

    clef

    Cloudflare·Vision

    ≈19 GB

    Clef is a 27B multimodal model that turns a state and a schema of typed questions into decisions. It reads the state as text, JSON, images, or video, and returns a probability for every allowed option of every question in a single forward pass.

    27.4B256K ctx Apache 2.0
    79KSept 2026ollama
    • Same category
    • Smaller (27.4B vs 219B)
    • Commercial use OK
    vs
  11. Qwen logo≈20 GB

    Meet Qwen3-VL — the most powerful vision-language model in the Qwen series to date.

    31.1BA3B256K ctx Apache 2.0
    418.4KSept 2025ollama
    • Same category
    • Smaller (31.1B vs 219B)
    • Commercial use OK
    vs
  12. meta-llama logo≈56 GB

    The Llama 3.2-Vision collection of multimodal large language models (LLMs) is a collection of pretrained and instruction-tuned image reasoning generative models in 11B and 90B sizes (text \+ images in / text out).

    88.6B128K ctx Llama 3.2 Community License
    425Sept 2024ollama
    • Same category
    • Smaller (88.6B vs 219B)
    vs
  13. openbmb logo

    MiniCPM-V-4.7-1B

    openbmb·Vision

    ≈1.8 GB

    MiniCPM-V 4.7 is the latest model in the MiniCPM-V series, available in a 1B variant. It builds on the family's single-image, multi-image, and video understanding capabilities, combining stronger foundation capabilities with efficient visual encoding and on-device inference.

    1.6B256K ctx Apache 2.0
    468.2KOct 2026gguf
    • Same category
    • Smaller (1.6B vs 219B)
    • Commercial use OK
    vs
  14. Qwen logo

    Qwen3.5-0.8B

    Qwen·Vision

    ≈1.4 GB

    Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance.

    873M256K ctx Apache 2.0
    2.6MFeb 2026ollama
    • Same category
    • Smaller (873M vs 219B)
    • Commercial use OK
    • Widely used
    vs
  15. HuggingFaceM4 logo

    idefics-80b-instruct

    HuggingFaceM4·Vision

    ≈53 GB

    How do I pronounce the model's name? Watch a Youtube tutorial

    80B2K ctx Custom licence
    8.4KJul 2023
    • Same category
    • Smaller (80B vs 219B)
    vs
  16. Qwen logo≈6.8 GB

    Meet Qwen3-VL — the most powerful vision-language model in the Qwen series to date.

    8.8B256K ctx Apache 2.0
    6MOct 2025ollama
    • Same category
    • Smaller (8.8B vs 219B)
    • Commercial use OK
    • Widely used
    vs
  17. google logo

    gemma-4-31B

    google·Vision

    ≈27 GB

    Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants.

    32.7B256K ctx Apache 2.0
    43.1KMar 2026gguf
    • Same category
    • Smaller (32.7B vs 219B)
    • Commercial use OK
    vs
  18. LGAI-EXAONE logo

    EXAONE-4.5-33B

    LGAI-EXAONE·Vision

    ≈23 GB

    We introduce EXAONE 4.5, the first open-weight vision language model developed by LG AI Research. Integrating a dedicated visual encoder into the existing EXAONE 4.0 framework, we expand the model's capability toward multimodality.

    34.4B256K ctx exaone
    23.8KApr 2026gguf
    • Same category
    • Smaller (34.4B vs 219B)
    vs
  19. moonshotai logo

    Kimi-K3

    moonshotai·Vision

    ≈1701 GB

    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.

    2.78T1M ctx kimi-k3
    734.8KJun 2026ollama
    • Same category
    • Larger (2.78T vs 219B)
    vs
  20. nvidia logo≈1.4 GB

    NVIDIA Nemotron Parse 2.0 transforms document images into structured, machine-readable representations with text, layout classes, bounding boxes, and reading-order information.

    903M9K ctx Custom licence
    589.3KJun 2026
    • Same category
    • Smaller (903M vs 219B)
    • Commercial use OK
    vs
  21. allenai logo

    Molmo-72B-0924

    allenai·Vision

    ≈45 GB

    Molmo is a family of open vision-language models developed by the Allen Institute for AI. Molmo models are trained on PixMo, a dataset of 1 million, highly-curated image-text pairs. It has state-of-the-art performance among multimodal models with a similar size while being fully open-source.

    73.3B4K ctx Apache 2.0
    1.3KSept 2024
    • Same category
    • Smaller (73.3B vs 219B)
    • Commercial use OK
    vs
  22. Qwen logo≈22 GB

    In addition to the original formula, we have further enhanced Qwen2.5-VL-32B's mathematical and problem-solving abilities through reinforcement learning.

    33.5B125K ctx Apache 2.0
    326.9KMar 2025ollama
    • Same category
    • Smaller (33.5B vs 219B)
    • Commercial use OK
    vs
  23. Qwen logo≈4.3 GB

    Meet Qwen3-VL — the most powerful vision-language model in the Qwen series to date.

    4.4B256K ctx Apache 2.0
    2.9MOct 2025ollama
    • Same category
    • Smaller (4.4B vs 219B)
    • Commercial use OK
    • Widely used
    vs
  24. Qwen logo

    Qwen3.5-0.8B-Base

    Qwen·Vision

    ≈1.4 GB

    Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance.

    873M256K ctx Apache 2.0
    881.9KFeb 2026gguf
    • Same category
    • Smaller (873M vs 219B)
    • Commercial use OK
    vs