Qwen3 VL 30B A3B Thinking
Qwen · released Sep 30, 2025 · Qwen/Qwen3-VL-30B-A3B-Thinking
- Type
- Open weightsApache License 2.0
- Params
- 31.1B
- Context
- 262K
3B active per word · about 197K words of context
Our take
Written Aug 3, 2026Qwen3 VL is a vision-language model from Alibaba that can be downloaded under a permissive Apache licence. It uses a mixture-of-experts design with only 3 billion active parameters per token from 31.1 billion total, and it handles text plus image input across a large context span.
Choose this for Apache-licensed image-to-text work where you need commercial freedom, or when you want to trade speed against cost on the same provider. The fastest tier we measured runs at 123 tokens per second. Skip it if you need benchmark-backed quality scores, if you want video input, or if you need consistent output pricing across providers.
The case for it
- Only 3 billion active parameters per token from 31.1 billion total — roughly a ten-to-one sparsity ratio.
- Apache 2.0 licence allows commercial use, fine-tuning and redistribution.
- 262,144-token context window for long multimodal documents.
- Same provider offers two speed tiers, letting you pick throughput against cost.
The case against it
- No benchmark scores in our data — no measured Elo, MMLU or other quality figures.
- Output pricing varies more than twofold across providers with no stated quality or speed justification.
- Throughput unmeasured for two of the five tracked offers.
How good is it?
We hold no score for this model.
We look for every model we track on every board we watch, and none of them has turned up Qwen3 VL 30B A3B Thinking — so there is no intelligence, coding, agentic or writing score to show you, not a low one, none.
Can you run it yourself?
- Fits in memory
- weights load entirely on the card
- Spills to system RAM
- some weights offload; much slower
- Too large
- will not load even with offload
- est
- size is calculated; the verdict could change by 10%
GeForce RTX 4090 · 24 GB
Borderline fit on an estimated size. It leaves 1.2 GB spare on a size we calculated rather than measured, and a 10% error either way would change the answer.
Comfortable fit
GeForce RTX 5090 · 32 GB
Room to spare. 9.2 GB spare means a 10% error in the size would not change the answer.
Apple M1 Pro (16-core GPU) · 32 GB
Room to spare. 2.4 GB spare means a 10% error in the size would not change the answer.
Memory use by level
Against a 24 GB card.
All 3 sizes here are calculated, not measured. We hold no measured file for this model, so each size comes from the parameter count and every verdict above inherits that uncertainty.
Check against your own machine → · All 71 devices, with every size →
Or rent it from someone else
Cheapest of 5 live listings. Picked at the widest standard context we hold, within one quantisation slice, so the numbers beside it are a price one host actually charges.
- per 1M tokens
- $0.20 in / $2.40 out
- Context served
- 262K
- Throughput
- Not measured
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| Novita AI | $0.20 / $1.00 | 131K | not measured | Unknown | Unknown | Unknown |
| SiliconFlowfp8 | $0.29 / $1.00 | 262K | 61 tok/s | No | No | Confirmed |
| Alibaba Cloud | $0.13 / $1.56 | 131K | 123 tok/s | No | Yesunknown period | Unknown |
| OpenRouter | $0.20 / $2.40 | 262K | not measured | Unknown | Unknown | Unknown |
| Alibaba Cloudfp8 | $0.20 / $2.40 | 131K | 125 tok/s | No | Yesunknown period | Unknown |
Novita AIAcross the 5 listings we hold: 3 say they do not train on prompts, 0 say they do and 2 do not say. 1 appear in the zero-retention registry we check; the rest are unknown to us rather than confirmed either way.
What each host's API supports
From the parameter list each endpoint publishes. Streaming is omitted: nothing we hold reports it, for any model.
- ✓
- Supported
- ✗
- Not supported
- Not published
- host gave no parameter list
| Provider | Tool calling | JSON output | Strict schema |
|---|---|---|---|
| Novita AI | |||
| SiliconFlowfp8 | ✓ | ✓ | ✓ |
| Alibaba Cloud | ✓ | ✓ | ✓ |
| OpenRouter | ✓ | ✓ | ✓ |
| Alibaba Cloudfp8 | ✓ | ✓ | ✓ |
Tool calling: 4 of 5 listings say yes, 1 publishes no parameter list. JSON output: 4 of 5 listings say yes, 1 publishes no parameter list. Strict schema: 4 of 5 listings say yes, 1 publishes no parameter list.
When we formed this view
Dates behind this page
Prices last checked 3d ago
What we do not know about this model yet
- We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
- No board we watch has turned up a score, so we hold no quality figures at all.
- 1 of 5 listings publish no parameter list, so what their API accepts is unknown to us.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- 2 of 5 listings do not say whether they train on prompts.
- We hold no cached-input rate for any of its listings.
Licence and identifiers
What the licence allowsApache License 2.0, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.
Licence
Apache License 2.0
Fully permissive: commercial use, redistribution, and derivatives allowed. Requires attribution and a copy of the license. Includes an express patent grant.
Identifiers
- Hugging Face
- Qwen/Qwen3-VL-30B-A3B-Thinking
- Architecture
- Mixture of experts
- Modality record
- text+image->text
- Catalogue slug
- qwen-qwen3-vl-30b-a3b-thinking