Up to date
Qwen Qwen License (qwen-research) out 21 months ago
3.75B parameters, images and text in, text out, 125K tokens of context. 1 provider serves it.
Verified 2026-10-25 Rank #77 in LLMs · score 103.4 method v1 Sources 3 Downloads 30d 1.7M · #76 of 619 tracked
Picked by rule from 29 tracked pieces: the lab's own post, the most shared, a field report, the latest.
The 2 picks above, plus these 23. 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; heavily run locally (98 quantized builds); a big base for builders (1,195 fine-tunes, adapters and merges).
Top Spaces running it: alexnasa/Wan2.2-Animate-ZEROGPU ♥ 365 · multimodalart/self-forcing ♥ 333 · alexnasa/Chain-of-Zoom ♥ 320 · fffiloni/Meigen-MultiTalk ♥ 276
Weights only, an estimate: leave room for the KV cache. Method
About 2.3 GB at INT4, 4.5 GB at FP8, 9.0 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 (2025-01-26), amber triangles are 6 dated articles (each links to it), grey ticks are repo commits.
1,293 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 | Qwen2_5_VLForConditionalGeneration |
| Model type | qwen2_5_vl |
| Dtype | bfloat16 |
| Params by dtype | BF16: 3,754,622,976 |
| Size band | M |
| MoE | no |
| Context length | 128,000 (config.json max_position_embeddings) |
| Papers | arXiv:2309.00071, arXiv:2409.12191, arXiv:2308.12966 |
| License | Qwen License (qwen-research) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · Qwen2_5_VLForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Qwen2_5_VLForConditionalGeneration · 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 Qwen2.5-VL-3B-Instruct: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 9.0 GB · Single GPU (<24 GB) |
|---|---|
| FP8 (1 byte/param) | 4.5 GB · Single GPU (<24 GB) |
| INT4 (0.5 bytes/param) | 2.3 GB · Edge (<4 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
quantized: 98finetune: 893adapter: 302
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| Qwen/Qwen2.5-VL-3B-Instruct-AWQ | quantized | AWQ | 3.75B | M | — | Qwen | 77 | 3.9M |
| nanonets/Nanonets-OCR2-3B | finetune | — | 3.75B | M | — | nanonets | 517 | 3.7M |
| nanonets/Nanonets-OCR-s | finetune | — | 3.75B | M | — | nanonets | 1.6k | 2.2M |
| typhoon-ai/typhoon-ocr-3b | finetune | — | 3.75B | M | apache-2.0 | typhoon-ai | 10 | 1.2M |
| RedHatAI/Qwen2.5-VL-3B-Instruct-quantized.w8a8 | quantized | -quant | 4.07B | M | apache-2.0 | RedHatAI | 2 | 576.3k |
| ChatDOC/OCRFlux-3B | finetune | — | 4.06B | M | — | ChatDOC | 370 | 523.7k |
| IDEA-Research/Rex-Omni | finetune | — | 4.07B | M | other | IDEA-Research | 66 | 373.1k |
| RedHatAI/Qwen2.5-VL-3B-Instruct-FP8-dynamic | quantized | -FP8 | 4.07B | M | apache-2.0 | RedHatAI | 3 | 330.9k |
| tsystems/colqwen2.5-3b-multilingual-v1.0 | adapter | — | — | — | mit | tsystems | 20 | 315.2k |
| ggml-org/Qwen2.5-VL-3B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | ggml-org | 17 | 295.3k |
| unsloth/Qwen2.5-VL-3B-Instruct | finetune | — | 3.75B | M | — | unsloth | 6 | 280.7k |
| unsloth/Qwen2.5-VL-3B-Instruct-unsloth-bnb-4bit | quantized | -bnb | 3.83B | M | — | unsloth | 16 | 256.1k |
| unsloth/Qwen2.5-VL-3B-Instruct-GGUF | quantized | GGUF | — | — | — | unsloth | 31 | 242.3k |
| lmstudio-community/Qwen2.5-VL-3B-Instruct-GGUF | quantized | GGUF | — | — | — | lmstudio-community | 0 | 171.9k |
| huihui-ai/Qwen2.5-VL-3B-Instruct-abliterated | quantized | GGUF | 3.75B | M | — | huihui-ai | 35 | 169.4k |
| remyxai/SpaceQwen2.5-VL-3B-Instruct | quantized | GGUF | 3.75B | M | — | remyxai | 18 | 154.7k |
| numind/NuExtract-2.0-4B | finetune | — | 3.75B | M | mit | numind | 30 | 153.1k |
| tsystems/colqwen2.5-3b-base | finetune | — | 3.75B | M | apache-2.0 | tsystems | 0 | 133.5k |
| RedHatAI/Qwen2.5-VL-3B-Instruct-quantized.w4a16 | quantized | -quant | 4.07B | M | apache-2.0 | RedHatAI | 4 | 130.9k |
| sherif1313/Arabic-handwritten-OCR-4bit-Qwen2.5-VL-3B-v2 | quantized | -4bit | 3.86B | M | apache-2.0 | sherif1313 | 15 | 127.6k |
| unsloth/Qwen2.5-VL-3B-Instruct-bnb-4bit | quantized | -bnb | 3.86B | M | — | unsloth | 5 | 127.3k |
| stanford-oval/churro-3B | finetune | — | 3.75B | M | other | stanford-oval | 20 | 114.6k |
| nomic-ai/colnomic-embed-multimodal-3b | adapter | — | — | — | — | nomic-ai | 41 | 112k |
| nomic-ai/nomic-embed-multimodal-3b | adapter | — | — | — | — | nomic-ai | 30 | 91.6k |
| convergence-ai/proxy-lite-3b | finetune | — | 3.75B | M | cc-by-nc-4.0 | convergence-ai | 146 | 54.8k |
| mlx-community/Qwen2.5-VL-3B-Instruct-4bit | finetune | -4bit | 3.75B | M | — | mlx-community | 8 | 60.5k |
| bytedance-research/Lance | finetune | — | — | — | apache-2.0 | bytedance-research | 1.1k | 6.4k |
| Vchitect/ShotVL-3B | finetune | — | 3.75B | M | apache-2.0 | Vchitect | 8 | 41.1k |
| mradermacher/Qwen2.5-VL-3B-Instruct-GGUF | quantized | GGUF | — | — | — | mradermacher | 1 | 38.1k |
| Hcompany/Holo1-3B | finetune | — | 3.75B | M | other | Hcompany | 81 | 25.8k |
| Diankun/Spatial-MLLM-subset-sft | finetune | — | 5.32B | M | mit | Diankun | 3 | 27.9k |
| OmniSVG/OmniSVG | finetune | — | — | — | apache-2.0 | OmniSVG | 209 | 16.5k |
| Metric-AI/ColQwen2.5-3b-multilingual-v1.0 | adapter | — | — | — | mit | Metric-AI | 10 | 26.3k |
| nvidia/NV-Reason-CXR-3B | finetune | — | 3.75B | M | other | nvidia | 36 | 21.5k |
| WaltonFuture/Qwen2.5VL-3b-RLCS | finetune | — | 4.07B | M | apache-2.0 | WaltonFuture | 0 | 20.8k |
| Metric-AI/colqwen2.5-3b-multilingual | adapter | — | — | — | mit | Metric-AI | 17 | 19.3k |
| microsoft/GUI-Actor-3B-Qwen2.5-VL | finetune | — | 3.79B | M | mit | microsoft | 14 | 17.9k |
| hfl/Qwen2.5-VL-3B-Instruct-GPTQ-Int4 | quantized | GPTQ | 3.75B | M | apache-2.0 | hfl | 3 | 17.5k |
| omlab/VLM-R1-Qwen2.5VL-3B-Math-0305 | finetune | — | 3.75B | M | apache-2.0 | omlab | 8 | 16.6k |
| sherif1313/Arabic-English-handwritten-OCR-v3 | finetune | — | 3.75B | M | apache-2.0 | sherif1313 | 17 | 15.4k |
| Date | Commit | SHA |
|---|---|---|
| 2025-04-06 | Update tokenizer_config.json | 6628554 |
| 2025-03-26 | Remove self referencing base model (#24) | c747f21 |
| 2025-03-23 | Update generation_config.json | 319ccfd |
| 2025-03-21 | Update generation_config.json | d16cc59 |
| 2025-02-15 | Update preprocessor_config.json | 1b989f2 |
| 2025-02-04 | Update README.md | b318d09 |
| 2025-01-27 | Update README.md | a72921a |
| 2025-01-27 | Update README.md | acbce9a |
| 2025-01-27 | Update README.md | 9c1a731 |
| 2025-01-27 | Create LICENSE | 37ce9f6 |
| 2025-01-27 | Create README.md | aa4cd35 |
| 2025-01-26 | Upload folder using huggingface_hub | 37eefda |
| 2025-01-26 | initial commit | 8e96c96 |
Hosted endpoints observed for Qwen2.5-VL-3B-Instruct, 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/Qwen2.5-VL-3B-Instruct | — | — | — | — | — | — | — | 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 21:33 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:00 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.
