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 ZwZ2-4B

120 open-weight models that do the same job as ZwZ2-4B (4.5B 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
inclusionAI logo

ZwZ2-4B

inclusionAI·Vision

≈4.2 GB

ZwZ2-4B is the Qwen3.5-4B-based model in Zooming-without-Zooming-v2 (ZwZ2). It learns to connect visual evidence across regions through comparison, aggregation, matching, and multi-step reasoning. Training uses full images; inference requires no zooming tools.

4.5B256K ctx Unknown licence
1.1KOct 2026gguf
--vram
any
--gpus
--quant
  1. LiquidAI logo

    d1-3B

    LiquidAI·Vision

    ≈2.8 GB

    d1-3B is a 3B parameter decision model built on LFM2.5-VL-3B. You give it a state (text, JSON, images, or a mix) and a set of questions. It returns calibrated, typed answers in one forward pass with zero output tokens.

    3.1B32K ctx lfm1.0
    19.1KOct 2026gguf
    • Same category
    • Similar size (3.1B vs 4.5B)
    vs
  2. GroundingPI logo

    GroundAnything-VLM

    GroundingPI·Vision

    ≈4.5 GB

    Model family: GroundAnything — DLM / parallel decoding · GroundAnything-VLM — autoregressive.

    4.8B256K ctx apache-2.0-with-upstream-kimi-k3
    1.2KSept 2026gguf
    • Same category
    • Similar size (4.8B vs 4.5B)
    • Commercial use OK
    vs
  3. allenai logo

    Molmo2-ER

    allenai·Vision

    ≈4.5 GB

    Molmo2-ER (Embodied Reasoning) is a 4B vision–language model specialized for the embodied perception skills that downstream action models depend on: scene understanding, pixel-accurate pointing, multi-image and egocentric–exocentric correspondence, and video temporal reasoning.

    4.9B16K ctx Apache 2.0
    3.5KMay 2026
    • Same category
    • Similar size (4.9B vs 4.5B)
    • Commercial use OK
    vs
  4. docling-project logo

    DeskForge-Qwen3.5-4B

    docling-project·Vision

    ≈4.2 GB

    DeskForge-Qwen3.5-4B is Qwen3.5-4B fine-tuned for desktop GUI grounding and next-action prediction using grounding and GUI-workflow supervision.

    4.5B256K ctx Apache 2.0
    623Sept 2026gguf
    • Same category
    • Similar size (4.5B vs 4.5B)
    • Commercial use OK
    vs
  5. google logo

    translategemma-4b-it

    google·Vision

    ≈4.5 GB

    TranslateGemma is a family of lightweight, state-of-the-art open translation models from Google, based on the Gemma 3 family of models. TranslateGemma models are designed to handle translation tasks across 55 languages.

    5B128K ctx Gemma Terms of Use
    10.5KJan 2026ollama
    • Same category
    • Similar size (5B vs 4.5B)
    vs
  6. OpenGVLab logo

    InternVL3_5-4B

    OpenGVLab·Vision

    ≈4.4 GB

    We introduce InternVL3.5, a new family of open-source multimodal models that significantly advances versatility, reasoning capability, and inference efficiency along the InternVL series.

    4.7B40K ctx Apache 2.0
    16.2KAug 2025gguf
    • Same category
    • Similar size (4.7B vs 4.5B)
    • Commercial use OK
    vs
  7. GroundingPI logo

    GroundingPI

    GroundingPI·Vision

    ≈4.5 GB

    Precise grounding matters. It specifies which object is the target and where that object is, even in clutter and for tiny objects, and it has to be fast enough for closed-loop control.

    4.8B256K ctx apache-2.0-with-upstream-kimi-k3
    956Sept 2026
    • Same category
    • Similar size (4.8B vs 4.5B)
    • Commercial use OK
    vs
  8. openbmb logo

    MiniCPM-V-4

    openbmb·Vision

    ≈3.2 GB

    A GPT-4V Level MLLM for Single Image, Multi Image and Video on Your Phone

    4.1B32K ctx Apache 2.0
    30.4KJul 2025gguf
    • Same category
    • Similar size (4.1B vs 4.5B)
    • Commercial use OK
    vs
  9. meta-models logo

    Muse-Glimmer-30B-assistant

    meta-models·Vision

    ≈2.2 GB

    Muse Glimmer is a 30-billion-parameter causal language model with a dedicated perception encoder, distilled from Muse Spark and purpose-built for autonomous agentic tasks on consumer hardware.

    2.6B128K ctx Apache 2.0
    90.3KAug 2026
    • Same category
    • Similar size (2.6B vs 4.5B)
    • Commercial use OK
    vs
  10. allenai logo

    MolmoWeb-4B

    allenai·Vision

    ≈4.5 GB

    Important Update! We made a few small but important updates to this HF/transformers-compatible checkpoint to ensure exact outputs to our native model checkpoint on March 29, 2026 ~6PM PST. If you downloaded this model checkpoint earlier than this time, we recommend re-downloading it.

    4.9B10K ctx Apache 2.0
    2.4KMar 2026
    • Same category
    • Similar size (4.9B vs 4.5B)
    • Commercial use OK
    vs
  11. microsoft logo

    Phi-3.5-vision-instruct

    microsoft·Vision

    ≈6.0 GB

    Phi-3.5-vision is a lightweight, state-of-the-art open multimodal model built upon datasets which include - synthetic data and filtered publicly available websites - with a focus on very high-quality, reasoning dense data both on text and vision.

    4.1B128K ctx MIT
    237.6KAug 2024
    • Same category
    • Similar size (4.1B vs 4.5B)
    • Commercial use OK
    vs
  12. OpenGVLab logo

    InternVL2_5-4B-MPO

    OpenGVLab·Vision

    ≈3.0 GB

    We introduce InternVL2.5-MPO, an advanced multimodal large language model (MLLM) series that demonstrates superior overall performance. This series builds upon InternVL2.5 and Mixed Preference Optimization.

    3.7B32K ctx MIT
    245.9KDec 2024gguf
    • Same category
    • Similar size (3.7B vs 4.5B)
    • Commercial use OK
    vs
  13. ibm-granite logo

    granite-4.0-3b-vision

    ibm-granite·Vision

    ≈3.5 GB

    Model Summary: Granite-4.0-3B-Vision is a vision-language model (VLM) designed for enterprise-grade document data extraction. It focuses on specialized, complex extraction tasks that ultracompact models often struggle with:

    4B128K ctx Apache 2.0
    3.3KMar 2026gguf
    • Same category
    • Similar size (4B vs 4.5B)
    • Commercial use OK
    vs
  14. Qwen logo≈5.8 GB

    In the past five months since Qwen2-VL’s release, numerous developers have built new models on the Qwen2-VL vision-language models, providing us with valuable feedback. During this period, we focused on building more useful vision-language models.

    8.3B125K ctx Apache 2.0
    5.2MJan 2025ollama
    • Same category
    • Similar size (8.3B vs 4.5B)
    • Commercial use OK
    • Widely used
    vs
  15. stepfun-ai logo

    GELab-Zero-4B-preview

    stepfun-ai·Vision

    ≈4.3 GB

    This model is part of the GELab-Zero project, which presents the Step-GUI Technical Report [Paper] [Project Page] [Code].

    4.4B256K ctx Apache 2.0
    2.3KNov 2025gguf
    • Same category
    • Similar size (4.4B vs 4.5B)
    • Commercial use OK
    vs
  16. nvidia logo

    Cosmos-Reason2-2B

    nvidia·Vision

    ≈2.5 GB

    NVIDIA Cosmos Reason 2 is an open, customizable, 2B-parameter reasoning vision language model (VLM) for physical AI and robotics that enables robots and vision AI agents to reason like humans, using prior knowledge, physics understanding and common sense to understand and act in the real world.

    2.4B256K ctx NVIDIA Open Model License
    475.6KDec 2025gguf
    • Same category
    • Similar size (2.4B vs 4.5B)
    • Commercial use OK
    vs
  17. inclusionAI logo

    ZwZ-4B

    inclusionAI·Vision

    ≈4.5 GB

    ZwZ-4B is a fine-grained multimodal perception model built upon Qwen3-VL-4B. It is trained using Region-to-Image Distillation (R2I) combined with reinforcement learning, enabling superior fine-grained visual understanding in a single forward pass — no inference-time zooming or tool calling…

    4.8B256K ctx Apache 2.0
    1.6KFeb 2026gguf
    • Same category
    • Similar size (4.8B vs 4.5B)
    • Commercial use OK
    vs
  18. Qwen logo

    Qwen3.5-2B-Base

    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
    321.5KFeb 2026gguf
    • Same category
    • Similar size (2.3B vs 4.5B)
    • Commercial use OK
    vs
  19. google logo

    shieldgemma-2-4b-it

    google·Vision

    ≈4.1 GB

    ShieldGemma 2 is a model trained on Gemma 3's 4B IT checkpoint for image safety classification across key categories that takes in images and outputs safety labels per policy.

    4.3B8K ctx Gemma Terms of Use
    12KMar 2025gguf
    • Same category
    • Similar size (4.3B vs 4.5B)
    vs
  20. Zyphra logo

    ZAYA1-VL-8B

    Zyphra·Vision

    ≈6.6 GB

    ZAYA1-VL-8B is a vision-language model (VLM) built upon Zyphra's ZAYA1-8B LLM. It has state-of-the-art performance among VLMs for its size and inference efficiency.

    9.7B32K ctx Apache 2.0
    339.4KApr 2026gguf
    • Same category
    • Larger (9.7B vs 4.5B)
    • Commercial use OK
    vs
  21. Qwen logo≈2.6 GB

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

    2.1B256K ctx Apache 2.0
    1.2MOct 2025ollama
    • Same category
    • Smaller (2.1B vs 4.5B)
    • Commercial use OK
    vs
  22. tencent logo

    WeVisOne-VL-2B

    tencent·Vision

    ≈2.8 GB

    WeVisOne-VL (WeChat Vision All-in-One) is a unified vision-language model developed by WeChat Vision, Tencent Inc. Built upon Qwen3-VL, it unifies object-level perception (open-vocabulary object detection, referring localization, object counting, object pointing, and OCR) and pixel-level perception…

    2.4B256K ctx Apache 2.0
    18.5KOct 2026gguf
    • Same category
    • Similar size (2.4B vs 4.5B)
    • Commercial use OK
    • Newer release
    vs
  23. Qwen logo

    Qwen3.5-9B-Base

    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
    266.8KFeb 2026gguf
    • Same category
    • Larger (9.7B vs 4.5B)
    • Commercial use OK
    vs
  24. CohereLabs logo

    North-Micro-Vision-Instruct

    CohereLabs·Vision

    ≈2.9 GB

    North Micro Vision Instruct is a 2.4B-parameter open-weight vision-language model with native-resolution image support, released under the Apache 2.0 license. It is designed as a compact foundation for prototyping, task-specific fine-tuning, and specialized multimodal applications.

    2.5B500K ctx Apache 2.0
    16.7KAug 2026
    • Same category
    • Similar size (2.5B vs 4.5B)
    • Commercial use OK
    vs