Up to date
Qwen Apache 2.0 out 12 months ago
4.44B parameters, images and text in, text out, 256K tokens of context. 1 provider serves it.
Verified 2026-10-25 Rank #314 in LLMs · score 82.9 method v1 Sources 3 Downloads 30d 20.2k · #455 of 619 tracked
Picked by rule from 17 tracked pieces: the lab's own post, the most shared, a field report, the latest.
OpenAI released gpt-oss-120b and gpt-oss-20b, their first open-weight models since GPT-2 in 2019.
The 4 picks above, plus these 13. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.
Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.
| # | Cheapest live row | In /1M | Out /1M | Context | Quant | Seen |
|---|---|---|---|---|---|---|
| 1 | Featherless | — | — | — | — | Oct 25 |
Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.
Served by 1 API provider; 31 quantized builds for local use.
Top Spaces running it: baohuynhbk14/Qwen3-VL-Demo ♥ 57 · nvidia/Cosmos3-Action-Viewer ♥ 46 · Slait/Caption_Qwen-vl ♥ 2 · nvidia/Cosmos3-Action-Viewer-Prerelease ♥ 1
Weights only, an estimate: leave room for the KV cache. Method
About 2.7 GB at INT4, 5.3 GB at FP8, 11 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
Weights, every dated article, and repo commits on one rail.
Teal is the weights (2025-10-11), amber triangles are 4 dated articles (each links to it), grey ticks are repo commits.
73 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Everything else, folded: the full tables, every source.
Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.
Listed in LLMs
| Pipeline tag | image-text-to-text |
|---|---|
| Library | transformers |
| Architectures | Qwen3VLForConditionalGeneration |
| Model type | qwen3_vl |
| Dtype | bfloat16 |
| Params by dtype | BF16: 4,437,815,808 |
| Size band | M |
| MoE | no |
| Context length | 262,144 (config.json max_position_embeddings) |
| Papers | arXiv:2505.09388, arXiv:2502.13923, arXiv:2409.12191, arXiv:2308.12966 |
| License | Apache 2.0 (apache-2.0) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · Qwen3VLForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Qwen3VLForConditionalGeneration · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed
vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.
Weights-only memory for Qwen3-VL-4B-Thinking: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 11 GB · Single GPU (<24 GB) |
|---|---|
| FP8 (1 byte/param) | 5.3 GB · Single GPU (<24 GB) |
| INT4 (0.5 bytes/param) | 2.7 GB · Edge (<4 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
quantized: 31finetune: 34adapter: 8
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| unsloth/Qwen3-VL-4B-Thinking-unsloth-bnb-4bit | quantized | -bnb | 4.52B | M | apache-2.0 | unsloth | 2 | 534.1k |
| unsloth/Qwen3-VL-4B-Thinking-GGUF | quantized | GGUF | — | — | apache-2.0 | unsloth | 26 | 176.4k |
| Qwen/Qwen3-VL-4B-Thinking-FP8 | quantized | -FP8 | 4.83B | M | apache-2.0 | Qwen | 39 | 93.2k |
| Hcompany/Holo2-4B | finetune | — | 4.44B | M | apache-2.0 | Hcompany | 26 | 75.4k |
| cyankiwi/Qwen3-VL-4B-Thinking-AWQ-4bit | quantized | AWQ | 4.83B | M | apache-2.0 | cyankiwi | 1 | 72.4k |
| Qwen/Qwen3-VL-4B-Thinking-GGUF | quantized | GGUF | — | — | apache-2.0 | Qwen | 25 | 50.9k |
| unsloth/Qwen3-VL-4B-Thinking-1M-GGUF | quantized | GGUF | — | — | apache-2.0 | unsloth | 5 | 22.5k |
| bartowski/Qwen_Qwen3-VL-4B-Thinking-GGUF | quantized | GGUF | — | — | — | bartowski | 3 | 16.1k |
| NexaAI/Qwen3-VL-4B-Thinking-GGUF | quantized | GGUF | — | — | — | NexaAI | 7 | 13.3k |
| huihui-ai/Huihui-Qwen3-VL-4B-Thinking-abliterated | finetune | — | 4.44B | M | apache-2.0 | huihui-ai | 29 | 11k |
| furiosa-ai/Qwen3-VL-4B-Thinking | finetune | — | 4.44B | M | apache-2.0 | furiosa-ai | 0 | 9.1k |
| ZuzeTt/Qwen3-VL-4B-Thinking-heretic-imatrix-GGUF | quantized | imatrix | — | — | apache-2.0 | ZuzeTt | 1 | 7.9k |
| mradermacher/Qwen3-VL-4B-Thinking-i1-GGUF | quantized | GGUF | — | — | apache-2.0 | mradermacher | 0 | 7.8k |
| prithivMLmods/Qwen3-VL-4B-Thinking-abliterated-v1 | finetune | — | 4.44B | M | apache-2.0 | prithivMLmods | 19 | 5.5k |
| unsloth/Qwen3-VL-4B-Thinking | finetune | — | 4.44B | M | apache-2.0 | unsloth | 2 | 6.3k |
| lmstudio-community/Qwen3-VL-4B-Thinking-GGUF | quantized | GGUF | — | — | apache-2.0 | lmstudio-community | 0 | 6.3k |
| TirGun/Qwen3-VL-4B-Thinking-GGUF | quantized | GGUF | — | — | apache-2.0 | TirGun | 1 | 4.6k |
| nvidia/EGM-4B | finetune | — | 4.83B | M | apache-2.0 | nvidia | 12 | 3.6k |
| unsloth/Qwen3-VL-4B-Thinking-bnb-4bit | quantized | -bnb | 4.55B | M | apache-2.0 | unsloth | 1 | 3.8k |
| taobao-mnn/Qwen3-VL-4B-Thinking-MNN | quantized | — | — | — | apache-2.0 | taobao-mnn | 3 | 3.3k |
| cccccccccc418/Qwen3-VL-4B-Thinking-GGUF | quantized | GGUF | — | — | apache-2.0 | cccccccccc418 | 0 | 3.3k |
| mradermacher/Qwen3-VL-4B-Thinking-GGUF | quantized | GGUF | — | — | apache-2.0 | mradermacher | 0 | 2.9k |
| nvidia/EGM-4B-SFT | finetune | — | 570K | S | apache-2.0 | nvidia | 1 | 2.4k |
| redponike/Qwen3-VL-4B-Thinking-GGUF | quantized | GGUF | — | — | apache-2.0 | redponike | 0 | 2.1k |
| DevQuasar/Qwen.Qwen3-VL-4B-Thinking-GGUF | quantized | GGUF | — | — | — | DevQuasar | 0 | 2k |
| issai/Qolda-AVL-5B | finetune | — | 5.11B | M | apache-2.0 | issai | 2 | 1.8k |
| John1604/Qwen3-VL-4B-Thinking-gguf | quantized | gguf | — | — | apache-2.0 | John1604 | 1 | 1.8k |
| lmstudio-community/Qwen3-VL-4B-Thinking-MLX-4bit | quantized | MLX | 4.44B | M | apache-2.0 | lmstudio-community | 0 | 1.4k |
| cybermotaz/qwen3-vl-4b-thinking-nvfp4-w4a16 | quantized | nvfp4 | 2.85B | M | apache-2.0 | cybermotaz | 1 | 1.3k |
| lmstudio-community/Qwen3-VL-4B-Thinking-MLX-8bit | quantized | MLX | 4.44B | M | apache-2.0 | lmstudio-community | 0 | 1k |
| cyankiwi/Qwen3-VL-4B-Thinking-AWQ-8bit | quantized | AWQ | 2.22B | M | apache-2.0 | cyankiwi | 0 | 783 |
| coder3101/Qwen3-VL-4B-Thinking-heretic | finetune | — | 4.44B | M | apache-2.0 | coder3101 | 0 | 675 |
| taobao-mnn/Qwen3-VL-4B-Thinking-Eagle3 | finetune | — | 218M | S | apache-2.0 | taobao-mnn | 1 | 495 |
| prithivMLmods/Qwen3-VL-4B-Thinking-Unredacted-MAX | finetune | — | 4.44B | M | apache-2.0 | prithivMLmods | 1 | 472 |
| lmstudio-community/Qwen3-VL-4B-Thinking-MLX-5bit | quantized | MLX | 4.44B | M | apache-2.0 | lmstudio-community | 0 | 477 |
| lmstudio-community/Qwen3-VL-4B-Thinking-MLX-6bit | quantized | MLX | 4.44B | M | apache-2.0 | lmstudio-community | 0 | 384 |
| Surpem/Supertron-VL-4B | finetune | — | 4.44B | M | apache-2.0 | Surpem | 3 | 231 |
| mjf-su/PhysicalAI-base-VLA | finetune | — | 4.44B | M | mit | mjf-su | 0 | 360 |
| Mater1984/ollama_Qwen3-VL-4B-Thinking-abliterated_v1 | finetune | — | 4.44B | M | apache-2.0 | Mater1984 | 0 | 330 |
| WariHima/qwen3vl-4b-ja-gptoss-r1-distil-pre | finetune | — | 4.44B | M | apache-2.0 | WariHima | 0 | 306 |
| Date | Commit | SHA |
|---|---|---|
| 2025-10-15 | Adding `transformers` as the library. (#3) | 1de27d8 |
| 2025-10-15 | Update README.md (#1) | 1efb70e |
| 2025-10-15 | Fix: move inputs to model device in inference example to avoid "same device" error (#2) | b099169 |
| 2025-10-15 | Update README.md | eb2af9a |
| 2025-10-14 | Update README.md | c7220a2 |
| 2025-10-13 | Update README.md | fb0141a |
| 2025-10-13 | Update README.md | aa49a10 |
| 2025-10-11 | Upload folder using huggingface_hub | 428f2bc |
| 2025-10-11 | initial commit | e62237c |
Hosted endpoints observed for Qwen3-VL-4B-Thinking, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| Featherless hf | Qwen/Qwen3-VL-4B-Thinking | — | — | — | — | — | — | — | live | 2026-10-25 |
Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.
arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:38 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:08 UTC · Last verified 2026-10-25
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.
