Up to date
Qwen Apache 2.0 out 13 months ago
31.1B parameters, 3B active (8 of 128 experts per token), images and text in, text out, 256K tokens of context. 4 providers serve it from $0.15 per million input tokens.
Verified 2026-10-25 Rank #48 in LLMs · score 108.3 method v1 Sources 4 Downloads 30d 303.5k · #216 of 619 tracked
Picked by rule from 19 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 3 picks above, plus these 16. 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.
2 providers with a live price, from $0.15 to $0.29 per million input tokens; the median is $0.22. One dot per provider, cheapest in teal.
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 4 API providers; heavily run locally (60 quantized builds).
Top Spaces running it: smolagents/computer-use-agent ♥ 163 · nvidia/Cosmos3-Action-Viewer ♥ 46 · Nymbo/Serverless-TextGen-Hub ♥ 44 · lerobot/steer-a-robot ♥ 25
Weights only, an estimate: leave room for the KV cache. Method
About 19 GB at INT4, 37 GB at FP8, 75 GB at BF16: weights only, so leave room for the KV cache. Every expert must be resident, so total parameters count.
Estimate params × bytes per param × 1.2 · how we estimate
Weights, every dated article, and repo commits on one rail.
Teal is the weights (2025-09-30), amber triangles are 5 dated articles (each links to it), grey ticks are repo commits.
103 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.
| Pipeline tag | image-text-to-text |
|---|---|
| Library | transformers |
| Architectures | Qwen3VLMoeForConditionalGeneration |
| Model type | qwen3_vl_moe |
| Dtype | bfloat16 |
| Params by dtype | BF16: 31,070,754,032 |
| Size band | L |
| MoE | yes |
| Experts | 128 |
| Experts per token | 8 |
| 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 · Qwen3VLMoeForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Qwen3VLMoeForConditionalGeneration · 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-30B-A3B-Instruct: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 75 GB · Single high-end GPU (<80 GB) |
|---|---|
| FP8 (1 byte/param) | 37 GB · Single high-end GPU (<80 GB) |
| INT4 (0.5 bytes/param) | 19 GB · Single GPU (<24 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra. MoE: all experts must be resident, so total parameters count.
quantized: 60finetune: 39adapter: 4
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| QuantTrio/Qwen3-VL-30B-A3B-Instruct-AWQ | quantized | AWQ | 31.1B | L | apache-2.0 | QuantTrio | 45 | 9.8M |
| Qwen/Qwen3-VL-30B-A3B-Instruct-FP8 | quantized | -FP8 | 31.1B | L | apache-2.0 | Qwen | 122 | 4.2M |
| Qwen/Qwen3-VL-30B-A3B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | Qwen | 34 | 993k |
| lmstudio-community/Qwen3-VL-30B-A3B-Instruct-MLX-4bit | quantized | MLX | — | — | apache-2.0 | lmstudio-community | 1 | 812.8k |
| unsloth/Qwen3-VL-30B-A3B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | unsloth | 121 | 753k |
| lmstudio-community/Qwen3-VL-30B-A3B-Instruct-MLX-8bit | quantized | MLX | — | — | apache-2.0 | lmstudio-community | 2 | 754k |
| lmstudio-community/Qwen3-VL-30B-A3B-Instruct-MLX-5bit | quantized | MLX | — | — | apache-2.0 | lmstudio-community | 0 | 737.6k |
| lmstudio-community/Qwen3-VL-30B-A3B-Instruct-MLX-6bit | quantized | MLX | — | — | apache-2.0 | lmstudio-community | 0 | 736.6k |
| cyankiwi/Qwen3-VL-30B-A3B-Instruct-AWQ-4bit | quantized | AWQ | 5.85B | M | apache-2.0 | cyankiwi | 10 | 550.1k |
| lmstudio-community/Qwen3-VL-30B-A3B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | lmstudio-community | 2 | 306.5k |
| browser-use/bu-30b-a3b-preview | finetune | — | 31.1B | L | — | browser-use | 269 | 233.5k |
| huihui-ai/Huihui-Qwen3-VL-30B-A3B-Instruct-abliterated | finetune | — | 31.1B | L | apache-2.0 | huihui-ai | 94 | 141.4k |
| bartowski/Qwen_Qwen3-VL-30B-A3B-Instruct-GGUF | quantized | GGUF | — | — | — | bartowski | 11 | 71.9k |
| ig1/Qwen3-VL-30B-A3B-Instruct-NVFP4 | quantized | NVFP4 | 18.0B | L | apache-2.0 | ig1 | 10 | 51.4k |
| unsloth/Qwen3-VL-30B-A3B-Instruct-1M-GGUF | quantized | GGUF | — | — | apache-2.0 | unsloth | 8 | 37.7k |
| Intel/Qwen3-VL-30B-A3B-Instruct-int4-AutoRound | quantized | int4 | 1.17B | S | — | Intel | 4 | 33.3k |
| prithivMLmods/Qwen3-VL-30B-A3B-Instruct-abliterated-v1 | finetune | — | 31.1B | L | apache-2.0 | prithivMLmods | 3 | 23.2k |
| unsloth/Qwen3-VL-30B-A3B-Instruct | finetune | — | 31.1B | L | apache-2.0 | unsloth | 3 | 21.5k |
| mradermacher/Qwen3-VL-30B-A3B-Instruct-i1-GGUF | quantized | GGUF | — | — | apache-2.0 | mradermacher | 0 | 16.5k |
| yairpatch/Qwen3-VL-30B-A3B-Instruct-GGUF | finetune | GGUF | — | — | apache-2.0 | yairpatch | 25 | 11.6k |
| Mungert/bu-30b-a3b-preview-GGUF | quantized | GGUF | — | — | — | Mungert | 0 | 12.4k |
| N8Programs/Unslopper-GGUF | quantized | GGUF | — | — | apache-2.0 | N8Programs | 20 | 6.4k |
| cyankiwi/Qwen3-VL-30B-A3B-Instruct-AWQ-8bit | quantized | AWQ | 9.58B | L | apache-2.0 | cyankiwi | 3 | 6.3k |
| mradermacher/Qwen3-VL-30B-A3B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | mradermacher | 0 | 6.1k |
| mlfoundations/Gelato-30B-A3B | finetune | — | 31.1B | L | apache-2.0 | mlfoundations | 34 | 4.4k |
| ggml-org/Qwen3-VL-30B-A3B-Instruct-Q8_0-GGUF | quantized | Q8_0 | — | — | apache-2.0 | ggml-org | 3 | 5.8k |
| noctrex/Qwen3-VL-30B-A3B-Instruct-1M-MXFP4_MOE-GGUF | quantized | GGUF | — | — | — | noctrex | 3 | 4.1k |
| N8Programs/Unslopper-30B-A3B-bf16 | finetune | -bf16 | 30.5B | L | apache-2.0 | N8Programs | 55 | 829 |
| DevQuasar/Qwen.Qwen3-VL-30B-A3B-Instruct-GGUF | quantized | GGUF | — | — | — | DevQuasar | 0 | 2.9k |
| Hugguf/Qwen3-VL-30B-A3B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | Hugguf | 2 | 2.6k |
| inference-optimization/Qwen3-VL-1.0B-A0.4B-Instruct | finetune | — | 1.43B | S | mit | inference-optimization | 1 | 2.5k |
| MBZUAI/MediX-R1-30B | finetune | — | 31.1B | L | cc-by-nc-sa-4.0 | MBZUAI | 5 | 2.1k |
| redponike/Qwen3-VL-30B-A3B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | redponike | 0 | 2.1k |
| EnderCrypt/Qwen3-VL-30B-A3B-Instruct-GGUF-mmproj | quantized | GGUF | — | — | apache-2.0 | EnderCrypt | 0 | 2k |
| tripolskypetr/Qwen3-VL-30B-A3B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | tripolskypetr | 0 | 2k |
| dark-side-of-the-code/Qwen3-VL-30B-A3B-Instruct-AWQ | quantized | AWQ | 5.14B | M | apache-2.0 | dark-side-of-the-code | 0 | 1.8k |
| cmh/Jan-v2-VL-max-Instruct-FP8-GGUF | quantized | -FP8 | — | — | apache-2.0 | cmh | 0 | 1.7k |
| lovedheart/Qwen3-VL-30B-A3B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | lovedheart | 1 | 1.5k |
| sbull-dell/Qwen3-VL-30B-A3B-Instruct-NVFP4 | quantized | NVFP4 | 16.1B | L | — | sbull-dell | 0 | 1.6k |
| sintanial/bu-30b-a3b-preview | finetune | — | 31.1B | L | — | sintanial | 0 | 1.4k |
Hosted endpoints observed for Qwen3-VL-30B-A3B-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: $0.15/M.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| DeepInfra hf, deepinfra, openrouter | Qwen/Qwen3-VL-30B-A3B-Instruct | 256K | fp8 | $0.15 | $0.60 | 25.2 | 335 ms | yes | live | 2026-10-25 |
| Novita hf, openrouter | qwen/qwen3-vl-30b-a3b-instruct | 128K | bf16 | $0.20 | $0.70 | 37.6 | 919 ms | yes | status -2 | 2026-10-25 |
| SiliconFlow openrouter | qwen/qwen3-vl-30b-a3b-instruct | 256K | fp8 | $0.29 | $1.00 | — | — | yes | live | 2026-10-24 |
| Featherless hf | Qwen/Qwen3-VL-30B-A3B-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 21:32 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.
