Qwen3 VL 235B A22B Instruct
Qwen · released Sep 22, 2025 · Qwen/Qwen3-VL-235B-A22B-Instruct
- Type
- Open weightsApache License 2.0
- Params
- 236B
- Context
- 262K
22B active per word · about 197K words of context
Our take
Written Aug 2, 2026Qwen3 VL is a vision-language model from Alibaba whose weights can be downloaded under a permissive Apache licence. It uses a mixture-of-experts design that keeps inference costs down, and its strongest measured skill is coding rather than creative writing.
Choose this for vision-language tasks where you need open weights with commercial freedom, or for coding workloads where it scores highest among its own capabilities. Pick it when you want to trade price against speed: the cheapest host is less than half the throughput of the fastest one. Skip it if creative writing quality is critical, or if you need measured math strength — its scores there are middle-of-pack even against itself.
The case for it
- Strongest measured skill is coding, with an Arena Elo 49.8 points above its overall text score.
- Only 22 billion of its 236 billion parameters activate per token — about one in eleven — which keeps inference costs at mid-size levels despite frontier-scale capacity.
- Apache 2.0 licence allows commercial use, fine-tuning and redistribution without attribution.
- Nine tracked hosts with a real price-versus-speed tradeoff: throughput ranges from 10 to 24 tokens per second.
The case against it
- Creative writing is its weakest measured category, 103.8 points below its coding score on the same leaderboard.
- Math sits in the middle of its own scores, 54.4 points below coding and barely above its overall text average.
- The cheapest host is 2.4 times slower than the fastest; you pay a premium in speed or in price.
How good is it?
IntelligencePuzzles, maths, exam questions
Arena Text (overall)64th of 143 · 1415.1
CodingWriting and fixing code on its own
Arena Coding59th of 143 · 1465
Arena Coding is the only board that has scored it for this.
AgenticPlanning, calling tools, staying on task
Nobody we watch has scored Qwen3 VL 235B A22B Instruct for this. We would take the rating from Arena Agent (IPS).
WritingWe do not rate this
Two boards come close and neither tests writing: Arena Creative Writing asks people which of two replies they prefer, and LiveBench Language tests whether a model understood a passage. So we show where Qwen3 VL 235B A22B Instruct placed and give it no mark out of five.
These tests check whether a model follows instructions — a precondition for all the work above, but not a measure of how well that work is done, which is why they get no rating.
Every published score for this model6 scoresEvery figure we hold, from 6 boards, with who ran it and a link to the source — including the boards no rating above is built on.
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
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Apple M1 Pro (16-core GPU) · 32 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Comfortable fit
Apple M3 Ultra (80-core GPU) · 512 GB
Room to spare. 229.3 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 9 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
- $0.21 in / $1.90 out
- Context served
- 262K
- Throughput
- Not measured
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| DeepInfrafp8 | $0.20 / $0.88 | 262K | 21 tok/s | No | No | Confirmed |
| DeepInfrafp8 | $0.20 / $0.88 | 262K | not measured | Unknown | Unknown | Unknown |
| Alibaba Cloud | $0.26 / $1.04 | 131K | 24 tok/s | No | Yesunknown period | Unknown |
| Alibaba Cloudfp8 | $0.26 / $1.04 | 131K | 36 tok/s | No | Yesunknown period | Unknown |
| Novita AI | $0.30 / $1.50 | 131K | not measured | Unknown | Unknown | Unknown |
| Novita AIbf16 | $0.30 / $1.50 | 131K | 16 tok/s | No | No | Confirmed |
| Parasailfp8 | $0.21 / $1.90 | 131K | 12 tok/s | No | No | Confirmed |
| Venice AIfp8 | $0.21 / $1.90 | 128K | 9 tok/s | No | No | Confirmed |
| OpenRouter | $0.21 / $1.90 | 262K | not measured | Unknown | Unknown | Unknown |
DeepInfra
Novita AIAcross the 9 listings we hold: 6 say they do not train on prompts, 0 say they do and 3 do not say. 4 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 |
|---|---|---|---|
| DeepInfrafp8 | ✓ | ✓ | ✓ |
| DeepInfrafp8 | |||
| Alibaba Cloud | ✓ | ✓ | ✓ |
| Alibaba Cloudfp8 | ✓ | ✓ | ✓ |
| Novita AI | |||
| Novita AIbf16 | ✓ | ✓ | ✓ |
| Parasailfp8 | ✓ | ✓ | ✓ |
| Venice AIfp8 | ✓ | ✓ | ✓ |
| OpenRouter | ✓ | ✓ | ✓ |
Tool calling: 7 of 9 listings say yes, 2 publish no parameter list. JSON output: 7 of 9 listings say yes, 2 publish no parameter list. Strict schema: 7 of 9 listings say yes, 2 publish no parameter list.
Models people weigh against Qwen3 VL 235B A22B Instruct
When we formed this view
Dates behind this page
Prices last checked 4d 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.
- 2 of 9 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.
- 3 of 9 listings do not say whether they train on prompts.
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-235B-A22B-Instruct
- Architecture
- Mixture of experts
- Modality record
- text+image->text
- Catalogue slug
- qwen-qwen3-vl-235b-a22b-instruct