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. 1 provider serves it from $0.29 per million input tokens.
Verified 2026-10-25 Rank #344 in LLMs · score 81.4 method v1 Sources 4 Downloads 30d 15.3k · #476 of 619 tracked
Picked by rule from 10 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 2 picks above, plus these 8. 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 | SiliconFlow | $0.29 | $1.00 | 256K | fp8 | Oct 24 |
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; 32 quantized builds for local use.
Top Spaces running it: nvidia/Cosmos3-Action-Viewer ♥ 46 · nvidia/Cosmos3-Action-Viewer-Prerelease ♥ 1
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 4 dated articles (each links to it), grey ticks are repo commits.
44 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-Thinking: 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: 32finetune: 12
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| Qwen/Qwen3-VL-30B-A3B-Thinking-FP8 | quantized | -FP8 | 31.1B | L | apache-2.0 | Qwen | 63 | 729.3k |
| QuantTrio/Qwen3-VL-30B-A3B-Thinking-AWQ | quantized | AWQ | 31.1B | L | apache-2.0 | QuantTrio | 15 | 314.8k |
| unsloth/Qwen3-VL-30B-A3B-Thinking-GGUF | quantized | GGUF | — | — | apache-2.0 | unsloth | 49 | 231.7k |
| Qwen/Qwen3-VL-30B-A3B-Thinking-GGUF | quantized | GGUF | — | — | apache-2.0 | Qwen | 18 | 138.5k |
| cyankiwi/Qwen3-VL-30B-A3B-Thinking-AWQ-4bit | quantized | AWQ | 5.85B | M | apache-2.0 | cyankiwi | 5 | 69k |
| bartowski/Qwen_Qwen3-VL-30B-A3B-Thinking-GGUF | quantized | GGUF | — | — | — | bartowski | 2 | 66.3k |
| unsloth/Qwen3-VL-30B-A3B-Thinking-1M-GGUF | quantized | GGUF | — | — | apache-2.0 | unsloth | 7 | 37.2k |
| janhq/Jan-v2-VL-max-gguf | quantized | gguf | — | — | apache-2.0 | janhq | 4 | 24.8k |
| yairpatch/Qwen3-VL-30B-A3B-Thinking-GGUF | finetune | GGUF | — | — | apache-2.0 | yairpatch | 29 | 16.1k |
| Mungert/Holo2-30B-A3B-GGUF | quantized | GGUF | — | — | cc-by-nc-4.0 | Mungert | 0 | 16.9k |
| lmstudio-community/Qwen3-VL-30B-A3B-Thinking-GGUF | quantized | GGUF | — | — | apache-2.0 | lmstudio-community | 0 | 14.9k |
| mradermacher/Qwen3-VL-30B-A3B-Thinking-i1-GGUF | quantized | GGUF | — | — | apache-2.0 | mradermacher | 1 | 12.7k |
| huihui-ai/Huihui-Qwen3-VL-30B-A3B-Thinking-abliterated | finetune | — | 31.1B | L | apache-2.0 | huihui-ai | 16 | 10.2k |
| standardmodelbio/smb-vision-v1 | finetune | — | 612M | S | cc-by-nc-4.0 | standardmodelbio | 0 | 10.7k |
| redponike/Qwen3-VL-30B-A3B-Thinking-GGUF | quantized | GGUF | — | — | apache-2.0 | redponike | 0 | 10.3k |
| Hcompany/Holo2-30B-A3B | finetune | — | 31.1B | L | cc-by-nc-4.0 | Hcompany | 51 | 7.5k |
| mradermacher/Qwen3-VL-30B-A3B-Thinking-GGUF | quantized | GGUF | — | — | apache-2.0 | mradermacher | 0 | 6k |
| unsloth/Qwen3-VL-30B-A3B-Thinking | finetune | — | 31.1B | L | apache-2.0 | unsloth | 5 | 4.7k |
| lmstudio-community/Qwen3-VL-30B-A3B-Thinking-MLX-8bit | quantized | MLX | — | — | apache-2.0 | lmstudio-community | 0 | 4k |
| OptimizeLLM/Qwen3-VL-30B-A3B-Thinking-NVFP4 | quantized | NVFP4 | 18.0B | L | apache-2.0 | OptimizeLLM | 2 | 3.6k |
| lmstudio-community/Qwen3-VL-30B-A3B-Thinking-MLX-4bit | quantized | MLX | — | — | apache-2.0 | lmstudio-community | 0 | 3.3k |
| janhq/Jan-v2-VL-max-FP8 | quantized | -FP8 | 31.1B | L | apache-2.0 | janhq | 29 | 1.5k |
| noctrex/Qwen3-VL-30B-A3B-Thinking-1M-MXFP4_MOE-GGUF | quantized | GGUF | — | — | — | noctrex | 3 | 2.8k |
| DevQuasar/Qwen.Qwen3-VL-30B-A3B-Thinking-GGUF | quantized | GGUF | — | — | — | DevQuasar | 0 | 2.9k |
| lmstudio-community/Qwen3-VL-30B-A3B-Thinking-MLX-6bit | quantized | MLX | — | — | apache-2.0 | lmstudio-community | 0 | 1.9k |
| cyankiwi/Qwen3-VL-30B-A3B-Thinking-AWQ-8bit | quantized | AWQ | 9.58B | L | apache-2.0 | cyankiwi | 4 | 1.7k |
| John1604/Qwen3-VL-30B-A3B-Thinking-gguf | quantized | gguf | — | — | apache-2.0 | John1604 | 0 | 1.6k |
| lmstudio-community/Qwen3-VL-30B-A3B-Thinking-MLX-5bit | quantized | MLX | — | — | apache-2.0 | lmstudio-community | 0 | 1.4k |
| taobao-mnn/Qwen3-VL-30B-A3B-Thinking-MNN | quantized | — | — | — | apache-2.0 | taobao-mnn | 1 | 1k |
| prefeitura-rio/Rio-3.1-Open-30B-VL | finetune | — | 31.1B | L | mit | prefeitura-rio | 2 | 580 |
| EmbeddedLLM/Qwen3-VL-30B-A3B-Thinking.w4a16 | quantized | — | 5.14B | M | apache-2.0 | EmbeddedLLM | 0 | 679 |
| prithivMLmods/Qwen3-VL-30B-A3B-Thinking-abliterated-v1 | finetune | — | 31.1B | L | apache-2.0 | prithivMLmods | 4 | 404 |
| catplusplus/Qwen3-VL-30B-A3B-Thinking-Heretic | finetune | — | 31.1B | L | apache-2.0 | catplusplus | 1 | 511 |
| navispace/Qwen3-VL-30B-A3B-Thinking-AWQ | quantized | AWQ | 5.14B | M | apache-2.0 | navispace | 0 | 361 |
| mistralrs-community/Qwen3-VL-30B-A3B-Thinking-UQFF | quantized | — | — | — | — | mistralrs-community | 0 | 324 |
| elichen-skymizer/Qwen3-VL-30B-A3B-Thinking-Q4_K_M-GGUF | quantized | GGUF | — | — | apache-2.0 | elichen-skymizer | 0 | 311 |
| elichen-skymizer/Qwen3-VL-30B-A3B-Thinking-Q3_K_M-GGUF | quantized | GGUF | — | — | apache-2.0 | elichen-skymizer | 0 | 268 |
| benbayibaurba/cardea-v0 | finetune | — | 31.1B | L | other | benbayibaurba | 1 | 114 |
| spotapovadm/Qwen3-VL-30B-A3B-Thinking-AWQ-4bit | quantized | AWQ | 5.85B | M | apache-2.0 | spotapovadm | 0 | 154 |
| nonoskygt/Qwen3-VL-30B-A3B-Thinking-abliterated-GPTQ-INT4 | quantized | GPTQ | 31.3B | L | apache-2.0 | nonoskygt | 0 | 113 |
Hosted endpoints observed for Qwen3-VL-30B-A3B-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. Cheapest observed input price: $0.29/M.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| SiliconFlow openrouter | qwen/qwen3-vl-30b-a3b-thinking | 256K | fp8 | $0.29 | $1.00 | — | — | yes | live | 2026-10-24 |
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.
| Title | Source | Published | First seen | Signals |
|---|---|---|---|---|
| Qwen3.5 122B and 35B models offer Sonnet 4.5 performance on local computers | venturebeat.com | 2026-02-28 | 2026-10-07 | hn |
| Qwen3-Coder-Next | qwen.ai official | 2026-02-03 | 2026-10-07 | hn |
| Qwen3-Max-Thinking | qwen.ai official | 2026-01-26 | 2026-10-07 | hn |
| Qwen3-TTS family is now open sourced: Voice design, clone, and generation | qwen.ai official | 2026-01-22 | 2026-10-07 | hn |
| Qwen3-VL | qwen.ai official | 2025-09-23 | 2026-10-07 | hn |
| Qwen3-Omni: Native Omni AI model for text, image and video | github.com | 2025-09-22 | 2026-10-07 | hn |
| Qwen3-Next | qwen.ai official | 2025-09-12 | 2026-10-07 | hn |
| GPT-OSS vs. Qwen3 and a detailed look how things evolved since GPT-2 | magazine.sebastianraschka.com | 2025-08-10 | 2026-10-07 | hn |
| Qwen3-Coder: Agentic coding in the world | qwenlm.github.io official | 2025-07-22 | 2026-10-07 | hn |
| Qwen3: Think deeper, act faster | qwenlm.github.io official | 2025-04-28 | 2026-10-07 | hn |
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 15:09 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.
