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-base

120 open-weight models that do the same job as Hy4-base (771B 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-base

tencent·Vision

≈459 GB

Hy4 is a new-generation Mixture-of-Experts (MoE) flagship model developed by the Tencent Hy Team, supporting text, image, and video understanding. The language backbone comprises 770B total parameters, of which 49B are activated per token.

771BA50.3B1M ctx tencent-hy-license
68Nov 2026
--vram
any
--gpus
--quant
  1. stepfun-ai logo

    step3

    stepfun-ai·Vision

    ≈221 GB

    Step3 is our cutting-edge multimodal reasoning model—built on a Mixture-of-Experts architecture with 321B total parameters and 38B active. It is designed end-to-end to minimize decoding costs while delivering top-tier performance in vision–language reasoning.

    321B64K ctx Apache 2.0
    48.6KJul 2025
    • Same category
    • Smaller (321B vs 771B)
    • Commercial use OK
    vs
  2. meta-llama logo≈240 GB

    The Llama 4 collection of models are natively multimodal AI models that enable text and multimodal experiences. These models leverage a mixture-of-experts architecture to offer industry-leading performance in text and image understanding.

    402BA17B1M ctx Llama 4 Community License
    13.9KApr 2025gguf
    • Same category
    • Similar size (402B vs 771B)
    vs
  3. dots-studio logo

    dots3-note-prev

    dots-studio·Vision

    ≈206 GB

    dots3-note preview is the first open-weight model in the dots3 family. It is a Mixture-of-Experts model with 280B total parameters, 16B activated parameters, and support for a context length of up to 512K tokens. The model can understand text, images, video, and audio, and produces text outputs.

    288BA25.1B512K ctx Apache 2.0
    4.2KAug 2026gguf
    • Same category
    • Smaller (288B vs 771B)
    • Commercial use OK
    vs
  4. Qwen logo≈142 GB

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

    236BA22B256K ctx Apache 2.0
    318KSept 2025ollama
    • Same category
    • Smaller (236B vs 771B)
    • Commercial use OK
    vs
  5. MiniMaxAI logo

    MiniMax-VL-01

    MiniMaxAI·Vision

    ≈274 GB

    We are delighted to introduce our MiniMax-VL-01 model. It adopts the "ViT-MLP-LLM" framework, which is a commonly used technique in the field of multimodal large language models.

    456BA48.8B8K ctx Unknown licence
    4.3KJan 2025
    • Same category
    • Similar size (456B vs 771B)
    vs
  6. cantina-security logo

    apex-flash-1-abliterated

    cantina-security·Vision

    ≈213 GB

    apex-flash-1-abliterated is an experimental derivative of apex-flash-1, Cantina Security's open-weights security model developed in partnership with Yeta (@yetalabs on X).

    321BA18.3B1M ctx MIT
    707Sept 2026gguf
    • Same category
    • Smaller (321B vs 771B)
    • Commercial use OK
    vs
  7. baidu logo≈253 GB

    The advanced capabilities of the ERNIE 4.5 models, particularly the MoE-based A47B and A3B series, are underpinned by several key technical innovations:

    424BA47B128K ctx Apache 2.0
    933Jun 2025
    • Same category
    • Similar size (424B vs 771B)
    • Commercial use OK
    vs
  8. Qwen logo≈142 GB

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

    236BA22B256K ctx Apache 2.0
    81.5KSept 2025gguf
    • Same category
    • Smaller (236B vs 771B)
    • Commercial use OK
    vs
  9. baidu logo≈253 GB

    The advanced capabilities of the ERNIE 4.5 models, particularly the MoE-based A47B and A3B series, are underpinned by several key technical innovations:

    424BA47B128K ctx Apache 2.0
    446Jun 2025
    • Same category
    • Similar size (424B vs 771B)
    • Commercial use OK
    vs
  10. CohereLabs logo

    command-a-plus-05-2026-bf16

    CohereLabs·Vision

    ≈131 GB

    Command A+ is an open source model with 25 billion active parameters and 218B total parameters model optimized for agentic, multilingual, and reasoning-heavy tasks with a focus on enterprise performance, while also providing support for vision inputs for processing image inputs.

    219BA25.5B200K ctx Apache 2.0
    14.3KMay 2026gguf
    • Same category
    • Smaller (219B vs 771B)
    • Commercial use OK
    vs
  11. baidu logo≈253 GB

    The advanced capabilities of the ERNIE 4.5 models, particularly the MoE-based A47B and A3B series, are underpinned by several key technical innovations:

    424BA47B128K ctx Apache 2.0
    256Jun 2025
    • Same category
    • Similar size (424B vs 771B)
    • Commercial use OK
    vs
  12. tencent logo

    Hy4-flash

    tencent·Vision

    ≈134 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.

    220B1M ctx tencent-hy-license
    2.1KNov 2026gguf
    • Same category
    • Smaller (220B vs 771B)
    vs
  13. The advanced capabilities of the ERNIE 4.5 models, particularly the MoE-based A47B and A3B series, are underpinned by several key technical innovations:

    424BA47B128K ctx Apache 2.0
    235Jun 2025
    • Same category
    • Similar size (424B vs 771B)
    • Commercial use OK
    vs
  14. internlm logo

    Intern-S1

    internlm·Vision

    ≈145 GB

    💻Github Repo • 🤗Model Collections • 📜Technical Report • 💬Online Chat

    241BA27.8B64K ctx Apache 2.0
    30KJul 2025gguf
    • Same category
    • Smaller (241B vs 771B)
    • Commercial use OK
    vs
  15. Cactus-Compute logo

    mirage

    Cactus-Compute·Vision

    ? GB

    The package downloads and caches the engine and weights on first use. This repository provides mirage.cact, mirage.safetensors and the model config.

    ?BApache 2.0
    43.7KNov 2026
    • Same category
    • Commercial use OK
    vs
  16. stepfun-ai logo

    Step-3.7-Flash

    stepfun-ai·Vision

    ≈131 GB

    Step 3.7 Flash is a 198B-parameter sparse Mixture-of-Experts (MoE) vision-language model that combines a 196B-parameter language backbone with a 1.8B-parameter vision encoder for native image understanding.

    201B256K ctx Apache 2.0
    9.6KMay 2026gguf
    • Same category
    • Smaller (201B vs 771B)
    • Commercial use OK
    vs
  17. meta-llama logo

    Llama-4-Maverick-17B-128E

    meta-llama·Vision

    ≈240 GB

    The Llama 4 collection of models are natively multimodal AI models that enable text and multimodal experiences. These models leverage a mixture-of-experts architecture to offer industry-leading performance in text and image understanding.

    402BA17B256K ctx Llama 4 Community License
    218Apr 2025gguf
    • Same category
    • Similar size (402B vs 771B)
    vs
  18. Qwen logo

    Qwen3.5-122B-A10B

    Qwen·Vision

    ≈75 GB

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

    125BA10B256K ctx Apache 2.0
    737.5KFeb 2026ollama
    • Same category
    • Smaller (125B vs 771B)
    • Commercial use OK
    vs
  19. 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
    489Nov 2026
    • Same category
    • Smaller (219B vs 771B)
    vs
  20. NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16 is the full-precision release of NVIDIA Nemotron 3.5 Super VL, a vision-language model for reasoning, coding, tool use, document understanding, and video understanding.

    124BA12B1M ctx openmdw-1.1
    84.9KOct 2026gguf
    • Same category
    • Smaller (124B vs 771B)
    • Commercial use OK
    vs
  21. PaddlePaddle logo

    UVDoc

    PaddlePaddle·Vision

    ? GB

    The main purpose of text image correction is to carry out geometric transformation on the image to correct the document distortion, inclination, perspective deformation and other problems in the image, so that the subsequent text recognition can be more accurate.

    ?BApache 2.0
    454.3KJun 2025
    • Same category
    • Commercial use OK
    vs
  22. PaddlePaddle logo

    PP-LCNet_x1_0_doc_ori

    PaddlePaddle·Vision

    ? GB

    The Document Image Orientation Classification Module is primarily designed to distinguish the orientation of document images and correct them through post-processing.

    ?BApache 2.0
    330.5KJun 2025
    • Same category
    • Commercial use OK
    vs
  23. Qwen logo

    Qwen3.8-27B

    Qwen·Vision

    ≈19 GB

    Following the widespread community adoption of the Qwen3.5 and Qwen3.6 series, we are pleased to introduce Qwen3.8, the most capable generation in the Qwen open-model family to date.

    27.8B256K ctx Apache 2.0
    6.6MAug 2026ollama
    • Same category
    • Smaller (27.8B vs 771B)
    • Commercial use OK
    • Widely used
    vs
  24. Agnes-AI logo

    Agnes-2.5-Pro-Theta

    Agnes-AI·Vision

    ≈117 GB

    Agnes 2.5 Pro Theta is a multimodal mixture-of-experts checkpoint distributed with its model implementation, tokenizer, chat template, and image/video processors. This release provides BF16 weights for local inference through Transformers.

    188BApache 2.0
    813Nov 2026
    • Same category
    • Smaller (188B vs 771B)
    • Commercial use OK
    vs