Up to date
Qwen Apache 2.0 out 13 months ago
236B parameters, 22B active (8 of 128 experts per token), images and text in, text out, 256K tokens of context. 3 providers serve it from $0.20 per million input tokens.
Verified 2026-10-25 Rank #103 in LLMs · score 99 method v1 Sources 5 Downloads 30d 290k · #222 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 $0.20 to $0.21 per million input tokens; the median is $0.21. 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 3 API providers; 25 quantized builds for local use.
Top Spaces running it: hadadxyz/ai ♥ 819 · umint/ai ♥ 258 · black-forest-labs/flux-3-image-klein ♥ 210 · smolagents/computer-use-agent ♥ 163
Weights only, an estimate: leave room for the KV cache. Method
About 141 GB at INT4, 283 GB at FP8, 566 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
Claim-matched tests, graded by code. Receipts, not a score.
{
"providers": [
{
"provider": "parasail",
"endpoint_model": "parasail-qwen3-vl-235b-a22b-instruct",
"source_url": "https://api.parasail.io/v1/chat/completions",
"ttft_ms_median": 677,
"ttft_any_ms_median": 677,
"tokens_per_s_median": 57.8,
"runs": [
{
"ttft_s": 0.677,
"ttft_any_s": 0.677,
"tokens_per_s": 57.8,
"latency_s": 2.91,
"tokens_out": 168,
"reasoning_chars": 0,
"finish": "stop"
},
{
"ttft_s": 0.527,
"ttft_any_s": 0.527,
"tokens_per_s": 63.2,
"latency_s": 2.66,
"tokens_out": 168,
"reasoning_chars": 0,
"finish": "stop"
},
{
"ttft_s": 0.711,
"ttft_any_s": 0.711,
"tokens_per_s": 52.3,
"latency_s": 3.04,
"tokens_out": 159,
"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 (2025-09-22), amber triangles are 6 dated articles (each links to it), grey ticks are repo commits.
39 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: 235,670,022,896 |
| Size band | XXL |
| 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-235B-A22B-Instruct: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 566 GB · Multi-GPU node (<640 GB) |
|---|---|
| FP8 (1 byte/param) | 283 GB · Multi-GPU node (<640 GB) |
| INT4 (0.5 bytes/param) | 141 GB · Multi-GPU node (<640 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: 25finetune: 14
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| Qwen/Qwen3-VL-235B-A22B-Instruct-FP8 | quantized | -FP8 | 236B | XXL | apache-2.0 | Qwen | 51 | 2M |
| QuantTrio/Qwen3-VL-235B-A22B-Instruct-AWQ | quantized | AWQ | 236B | XXL | apache-2.0 | QuantTrio | 14 | 92.3k |
| unsloth/Qwen3-VL-235B-A22B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | unsloth | 24 | 89.4k |
| RedHatAI/Qwen3-VL-235B-A22B-Instruct-NVFP4 | quantized | NVFP4 | 133B | XXL | apache-2.0 | RedHatAI | 15 | 87.4k |
| RedHatAI/Qwen3-VL-235B-A22B-Instruct-FP8-dynamic | quantized | -FP8 | 236B | XXL | apache-2.0 | RedHatAI | 4 | 72.7k |
| tiny-random/qwen3-vl-moe | finetune | — | 3.16M | S | — | tiny-random | 0 | 52.6k |
| nvidia/Qwen3-VL-235B-A22B-Instruct-NVFP4 | quantized | NVFP4 | 119B | XXL | — | nvidia | 9 | 29.6k |
| Qwen/Qwen3-VL-235B-A22B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | Qwen | 19 | 28.9k |
| unsloth/Qwen3-VL-235B-A22B-Instruct-1M-GGUF | quantized | GGUF | — | — | apache-2.0 | unsloth | 3 | 23.1k |
| DevQuasar/Qwen.Qwen3-VL-235B-A22B-Instruct-GGUF | quantized | GGUF | — | — | — | DevQuasar | 0 | 8.4k |
| noctrex/Qwen3-VL-235B-A22B-Instruct-1M-MXFP4_MOE-GGUF | quantized | GGUF | — | — | — | noctrex | 4 | 3.2k |
| lmstudio-community/Qwen3-VL-235B-A22B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | lmstudio-community | 0 | 3k |
| RedHatAI/Qwen3-VL-235B-A22B-Instruct-speculator.eagle3 | finetune | — | 1.20B | S | apache-2.0 | RedHatAI | 2 | 2.5k |
| zenlm/zen-designer-gguf | quantized | gguf | — | — | apache-2.0 | zenlm | 1 | 2k |
| QuantTrio/Qwen3-VL-235B-A22B-Instruct-FP8 | quantized | -FP8 | — | — | apache-2.0 | QuantTrio | 0 | 1.9k |
| RedHatAI/Qwen3-VL-235B-A22B-Instruct-FP8-block | quantized | -FP8 | 236B | XXL | apache-2.0 | RedHatAI | 3 | 1.7k |
| FiditeNemini/Qwen3-VL-235B-A22B-Instruct-Abliterated-3bit | finetune | — | 1.31K | S | apache-2.0 | FiditeNemini | 0 | 1.6k |
| Nekotekina/Qwen3-VL-235B-A22B-Instruct-Projected-Abliterated-GGUF | quantized | GGUF | — | — | apache-2.0 | Nekotekina | 2 | 1.5k |
| LibraxisAI/Qwen3-VL-235B-A22B-Instruct-mlx-nvfp4 | quantized | mlx | 236B | XXL | apache-2.0 | LibraxisAI | 0 | 1.5k |
| yujiepan/qwen3-vl-moe-tiny-random | finetune | — | 3.16M | S | — | yujiepan | 0 | 1.4k |
| unsloth/Qwen3-VL-235B-A22B-Instruct | finetune | — | 236B | XXL | apache-2.0 | unsloth | 2 | 1.1k |
| LibraxisAI/Qwen3-VL-235B-A22B-Instruct-mlx-mxfp8 | quantized | mlx | 236B | XXL | apache-2.0 | LibraxisAI | 1 | 1.1k |
| OpenMOSE/Qwen3-VL-REAP-145B-A22B | finetune | — | 145B | XXL | apache-2.0 | OpenMOSE | 8 | 726 |
| LibraxisAI/Qwen3-VL-235B-A22B-Instruct-mlx-mxfp4 | quantized | mlx | 236B | XXL | apache-2.0 | LibraxisAI | 0 | 1k |
| zenlm/zen-designer-235b-a22b-instruct | finetune | — | 236B | XXL | apache-2.0 | zenlm | 1 | 898 |
| KaraKaraWarehouse/Heretic-Qwen3-VL-235B-A22B-Instruct | finetune | — | 236B | XXL | apache-2.0 | KaraKaraWarehouse | 9 | 424 |
| mradermacher/Qwen3-VL-235B-A22B-Instruct-i1-GGUF | quantized | GGUF | — | — | apache-2.0 | mradermacher | 1 | 691 |
| GadflyII/Qwen3-VL-235B-A22B-Instruct-NVFP4 | quantized | NVFP4 | 133B | XXL | apache-2.0 | GadflyII | 0 | 673 |
| mradermacher/Qwen3-VL-235B-A22B-Instruct-GGUF | quantized | GGUF | — | — | apache-2.0 | mradermacher | 0 | 583 |
| mncai/hunmin_vlm_235b_v0.11_merged_cua | finetune | — | 236B | XXL | apache-2.0 | mncai | 3 | 303 |
| shivak/Qwen3-VL-235B-A22B-Instruct-W4A16 | quantized | -W4A16 | 38.7B | XL | apache-2.0 | shivak | 0 | 263 |
| numinousmuses/laguna-s-2.1-vision | finetune | — | — | — | mit | numinousmuses | 1 | 136 |
| timm/qwen3_vit_416m.qwen3_vl_235b_a22b | finetune | — | 415M | S | apache-2.0 | timm | 0 | 103 |
| gsting/Qwen3-VL-235B-A22B-Instruct-AWQ | quantized | AWQ | 236B | XXL | apache-2.0 | gsting | 0 | 102 |
| timm/qwen3_vit_416m_merge.qwen3_vl_235b_a22b | finetune | — | 455M | S | apache-2.0 | timm | 0 | 88 |
| amd/Qwen3-VL-235B-A22B-Instruct-ptpc | quantized | — | 236B | XXL | apache-2.0 | amd | 0 | 73 |
| Arsh9210/Qwen3-VL-235B-A22B-Instruct-NVFP4 | quantized | NVFP4 | 119B | XXL | — | Arsh9210 | 0 | 52 |
| timm/qwen3_vit_416m_enc.qwen3_vl_235b_a22b | finetune | — | 455M | S | apache-2.0 | timm | 0 | 41 |
| nm-testing/Qwen3-VL-235B-A22B-Instruct-FP8-BLOCK | finetune | -FP8 | — | — | apache-2.0 | nm-testing | 0 | 35 |
| Date | Commit | SHA |
|---|---|---|
| 2025-11-26 | Adding `transformers` as the library name (#14) | 710c138 |
| 2025-10-04 | Update README.md | 326b2e3 |
| 2025-09-23 | Update README.md | a85c315 |
| 2025-09-23 | Update README.md | cbaeae4 |
| 2025-09-23 | Update README.md | 3618751 |
| 2025-09-23 | Update README.md | 9a1ef91 |
| 2025-09-23 | Add files using upload-large-folder tool | 5a94164 |
| 2025-09-23 | Add files using upload-large-folder tool | 835c92b |
| 2025-09-23 | Add files using upload-large-folder tool | 462155f |
| 2025-09-23 | Add files using upload-large-folder tool | ece8947 |
| 2025-09-23 | Add files using upload-large-folder tool | 983c162 |
| 2025-09-23 | Add files using upload-large-folder tool | 14ea127 |
| 2025-09-22 | initial commit | 07ca860 |
Hosted endpoints observed for Qwen3-VL-235B-A22B-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: $0.20/M.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| DeepInfra hf, deepinfra, openrouter | Qwen/Qwen3-VL-235B-A22B-Instruct | 256K | fp8 | $0.20 | $0.88 | 7.8 | 770 ms | yes | live | 2026-10-25 |
| Parasail parasail, openrouter | parasail-qwen3-vl-235b-a22b-instruct | 128K | fp8 | $0.21 | $1.90 | — | — | yes | serverless | 2026-10-24 |
| Venice openrouter | qwen/qwen3-vl-235b-a22b-instruct | 125K | fp8 | $0.21 | $1.90 | — | — | yes | status -2 | 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.
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.