Up to date
Qwen Apache 2.0 out 12 months ago
8.77B parameters, images and text in, text out, 256K tokens of context. 2 providers serve it from $0.25 per million input tokens.
Verified 2026-10-25 Rank #20 in LLMs · score 118.9 method v1 Sources 5 Downloads 30d 5.2M · #30 of 619 tracked
Changed this week: 3 new quantized builds on the Hub (119 in all).2026-10-24 · 2 new quantized builds on the Hub (118 in all).2026-10-24
Picked by rule from 18 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 14. 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.
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 2 API providers; heavily run locally (119 quantized builds); a big base for builders (904 fine-tunes, adapters and merges).
Top Spaces running it: prithivMLmods/Qwen3-VL-Outpost ♥ 208 · embedl/hfviewer ♥ 181 · smolagents/computer-use-agent ♥ 163 · artificialguybr/qwen-vl ♥ 126
Weights only, an estimate: leave room for the KV cache. Method
About 5.3 GB at INT4, 11 GB at FP8, 21 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
Claim-matched tests, graded by code. Receipts, not a score.
{
"providers": [
{
"provider": "parasail",
"endpoint_model": "parasail-qwen3vl-8b-instruct",
"source_url": "https://api.parasail.io/v1/chat/completions",
"ttft_ms_median": 900,
"ttft_any_ms_median": 900,
"tokens_per_s_median": 70.2,
"runs": [
{
"ttft_s": 0.9,
"ttft_any_s": 0.9,
"tokens_per_s": 70.2,
"latency_s": 2.37,
"tokens_out": 166,
"reasoning_chars": 0,
"finish": "stop"
},
{
"ttft_s": 0.906,
"ttft_any_s": 0.906,
"tokens_per_s": 73.6,
"latency_s": 2.26,
"tokens_out": 166,
"reasoning_chars": 0,
"finish": "stop"
},
{
"ttft_s": 0.64,
"ttft_any_s": 0.64,
"tokens_per_s": 41.8,
"latency_s": 3.97,
"tokens_out": 166,
"reasoning_chars": 0,
"finish": "stop"
}
]
}
],
"note": "Single client, wall clock; tok/s counts every output token including hidden reasoning."
} 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.
1,023 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: 8,767,123,696 |
| 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-8B-Instruct: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 21 GB · Single GPU (<24 GB) |
|---|---|
| FP8 (1 byte/param) | 11 GB · Single GPU (<24 GB) |
| INT4 (0.5 bytes/param) | 5.3 GB · Single GPU (<24 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
quantized: 119finetune: 674adapter: 226merge: 4
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| Qwen/Qwen3-VL-8B-Instruct-FP8 | quantized | -FP8 | 8.77B | M | apache-2.0 | Qwen | 94 | 12.4M |
| OneThink/OneThinker-SFT-Qwen3-8B | finetune | — | 770K | S | apache-2.0 | OneThink | 5 | 3.4M |
| cyankiwi/Qwen3-VL-8B-Instruct-AWQ-4bit | quantized | AWQ | 2.91B | M | apache-2.0 | cyankiwi | 19 | 2M |
| Qwen/Qwen3-VL-Reranker-8B | finetune | — | 8.77B | M | apache-2.0 | Qwen | 181 | 2M |
| Qwen/Qwen3-VL-8B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | Qwen | 186 | 1.8M |
| lmstudio-community/Qwen3-VL-8B-Instruct-MLX-4bit | quantized | MLX | — | — | apache-2.0 | lmstudio-community | 4 | 1.6M |
| lmstudio-community/Qwen3-VL-8B-Instruct-MLX-5bit | quantized | MLX | — | — | apache-2.0 | lmstudio-community | 1 | 1.6M |
| lmstudio-community/Qwen3-VL-8B-Instruct-MLX-8bit | quantized | MLX | — | — | apache-2.0 | lmstudio-community | 6 | 1.5M |
| lmstudio-community/Qwen3-VL-8B-Instruct-MLX-6bit | quantized | MLX | — | — | apache-2.0 | lmstudio-community | 0 | 1.5M |
| unsloth/Qwen3-VL-8B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | unsloth | 99 | 815.5k |
| lmstudio-community/Qwen3-VL-8B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | lmstudio-community | 7 | 795k |
| sukhrobnurali/qwen3vl-resume-parser | finetune | — | 8.77B | M | apache-2.0 | sukhrobnurali | 4 | 623.9k |
| AtomicChat/Qwen-Image-2.1-Turbo-Uncensored-GGUF | finetune | GGUF | — | — | apache-2.0 | AtomicChat | 620 | 493.2k |
| JEILDLWLRMA/Qwen3-VL-8B-Instruct-NVFP4 | quantized | NVFP4 | 5.73B | M | apache-2.0 | JEILDLWLRMA | 4 | 507.5k |
| unsloth/Qwen3-VL-8B-Instruct-unsloth-bnb-4bit | quantized | -bnb | 8.77B | M | apache-2.0 | unsloth | 22 | 491.1k |
| unsloth/Qwen3-VL-8B-Instruct | finetune | — | 8.77B | M | apache-2.0 | unsloth | 10 | 331.2k |
| huihui-ai/Huihui-Qwen3-VL-8B-Instruct-abliterated | finetune | — | 8.77B | M | apache-2.0 | huihui-ai | 220 | 314.8k |
| unsloth/Qwen3-VL-8B-Instruct-bnb-4bit | quantized | -bnb | 8.77B | M | apache-2.0 | unsloth | 4 | 140k |
| GitMylo/nsfwvision-qwen3-vl-8b-v3-gguf | quantized | gguf | — | — | — | GitMylo | 38 | 121.4k |
| bartowski/Qwen_Qwen3-VL-8B-Instruct-GGUF | quantized | GGUF | — | — | — | bartowski | 11 | 121k |
| RLWRLD/RLDX-1-VLM | finetune | — | 8.78B | M | other | RLWRLD | 2 | 118.8k |
| SherlockID365/Qwen3-VL-8B-Instruct-quantized.w4a16 | quantized | -quant | 2.74B | M | apache-2.0 | SherlockID365 | 1 | 96.2k |
| TomoroAI/tomoro-colqwen3-embed-8b | finetune | — | 8.77B | M | apache-2.0 | TomoroAI | 32 | 63.1k |
| cyankiwi/Qwen3-VL-8B-Instruct-AWQ-8bit | quantized | AWQ | 3.77B | M | apache-2.0 | cyankiwi | 4 | 58.3k |
| taobao-mnn/Qwen3-VL-8B-Instruct-Eagle3 | finetune | — | 400M | S | apache-2.0 | taobao-mnn | 2 | 45.3k |
| aisingapore/Qwen-SEA-LION-v4-8B-VL | finetune | — | 8.77B | M | mit | aisingapore | 9 | 42.3k |
| unsloth/Qwen3-VL-8B-Instruct-1M-GGUF | quantized | GGUF | — | — | apache-2.0 | unsloth | 7 | 38.2k |
| TencentARC/TimeLens-8B | finetune | — | 8.77B | M | other | TencentARC | 21 | 31.3k |
| TechnoBaptist/Qwen-Image-2.1-Turbo-Uncensored-GGUF | finetune | GGUF | — | — | apache-2.0 | TechnoBaptist | 0 | 28.9k |
| kaitchup/Qwen3-VL-8B-Instruct-W4A16 | quantized | -W4A16 | 1.82B | S | apache-2.0 | kaitchup | 0 | 28.3k |
| prithivMLmods/Qwen3-VL-8B-Instruct-abliterated-v1 | finetune | — | 8.77B | M | apache-2.0 | prithivMLmods | 17 | 25.1k |
| GitMylo/nsfwcaption-qwen3-vl-8b-v3-gguf | quantized | gguf | — | — | — | GitMylo | 13 | 21.2k |
| DreamFast/Qwen3-VL-8B-Heretic-1.3.0 | finetune | GGUF | 8.77B | M | apache-2.0 | DreamFast | 36 | 19.7k |
| inclusionAI/ZwZ-8B | finetune | — | 8.77B | M | apache-2.0 | inclusionAI | 50 | 15.6k |
| prithivMLmods/Qwen3-VL-8B-Instruct-abliterated-v2 | finetune | — | 8.77B | M | apache-2.0 | prithivMLmods | 35 | 15.6k |
| hltcoe/RankVideo | finetune | — | 8.77B | M | mit | hltcoe | 1 | 16.3k |
| mradermacher/Qwen3-VL-8B-Instruct-i1-GGUF | quantized | GGUF | — | — | apache-2.0 | mradermacher | 1 | 16k |
| amd/Qwen3-VL-8B-Instruct-w8a8-llmcompressor | quantized | -w8a8 | 8.77B | M | apache-2.0 | amd | 0 | 14.9k |
| pottokao/Qwen-Image-2.1-Text-Encoder-Heretic | finetune | — | 8.77B | M | apache-2.0 | pottokao | 78 | 10.8k |
| wareef/Qwen3-VL-8B-Instruct-Q4_K_M-GGUF | quantized | GGUF | — | — | apache-2.0 | wareef | 5 | 13.8k |
| Date | Commit | SHA |
|---|---|---|
| 2025-10-15 | Adding `transformers` as the library name (#4) | 0c351dd |
| 2025-10-15 | Fix: move inputs to model device in inference example to avoid "same device" error (#3) | b7e6648 |
| 2025-10-15 | Update README.md (#2) | e0a319f |
| 2025-10-15 | Update readme to fix processor = AutoProcessor.from_pretrained("Qwen/Qwen/Qwen3-VL-8B-Instruct") to processor = AutoProcessor.from_pretrained("Qwen/Qwen3-VL-8B-Instruct") (#1) | 1447b2a |
| 2025-10-15 | Update README.md | 1dd1e02 |
| 2025-10-14 | Update README.md | cadac78 |
| 2025-10-13 | Update README.md | 434824e |
| 2025-10-11 | Upload folder using huggingface_hub | fe33404 |
| 2025-10-11 | initial commit | e766c28 |
Hosted endpoints observed for Qwen3-VL-8B-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. Cheapest observed input price: $0.25/M.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| Parasail parasail, openrouter | parasail-qwen3vl-8b-instruct | 256K | bf16 | $0.25 | $0.75 | — | — | yes | serverless | 2026-10-24 |
| Featherless hf | Qwen/Qwen3-VL-8B-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 21:31 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 14:59 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.