Qwen3 235B A22B Thinking 2507
Qwen · released Jul 25, 2025 · Qwen/Qwen3-235B-A22B-Thinking-2507
- Type
- Open weightsApache License 2.0
- Params
- 235B
- Context
- 262K
22B active per word · about 197K words of context
Our take
Written Sep 2, 2026Qwen3 is a large downloadable reasoning model from Alibaba with a permissive Apache licence. It activates 22 billion of its 235.1 billion parameters per token, and handles up to 262,144 tokens in a single request. Its coding score is its strongest measured skill, though output costs vary sharply by provider.
Pick this for coding tasks where its measured score is highest, hard-prompt work, or long-document jobs needing a quarter-million tokens under an open licence. Use it if you want commercial freedom without licensing restrictions. Skip it if creative writing quality matters most, or if you need predictable speed across providers.
The case for it
- Strongest measured skill is coding, 72 points above its own creative-writing score.
- Apache 2.0 licence permits commercial use, modification and redistribution.
- Activates only 22 billion parameters per token from 235.1 billion total — selective activation at very large scale.
- 262,144-token request limit suits long-document tasks.
The case against it
- Creative writing is its weakest measured skill, 72 points below coding and 47 below hard prompts.
- Throughput varies more than threefold by provider, from 62 to 18.5 tokens per second measured.
- Output rate is ten times the input rate on the cheapest tier, and some hosts charge more for slower service.
How good is it?
EverydayGeneral questions and everyday reasoning
Arena Text (overall)96th of 168 · 1400
CodingWriting and fixing code on its own
Arena Coding99th of 168 · 1442
Arena Coding is the only board that has scored it for this.
AgenticPlanning, calling tools, staying on task
Not yet scored on Arena Agent.
WritingDrafting and rewriting prose
Arena Creative Writing90th of 168 · 1370
Arena Creative Writing is the only board that has scored it for this.
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.
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?
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. 229.7 GB spare means a 10% error in the size would not change the answer.
Memory use by level
Against a 24 GB card.
What is quantisation? →This model on every device we track71 devicesThe Q4 build most people download, on each device: what the weights come to, how much context the memory leaves, and whether it runs. Smallest device that runs it first.
Check against your own machine → · Where to rent it hosted →
Or rent it from someone else
Prices checked between 1 hour and 46 days ago — each listing carries its own date.
- per 1M tokens
- $0.23 in / $2.30 out
- Context served
- 262K
- Throughput
- Not measured
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| OpenRouterOpenRouter's own listing | $0.23 / $2.30checked 1 hour ago | 131K | not measured | Unknown | Unknown | Unknown |
| DeepInfrafp8Direct | $0.23 / $2.30checked 46 days ago | 262K | not measured | Unknown | Unknown | Unknown |
| Alibaba CloudThrough OpenRouter | $0.23 / $2.30checked 1 hour ago | 131K118K max reply | 54 tok/s | No | Yesunknown period | Unknown |
| Novita AIfp8Direct and through OpenRouter | $0.30 / $3.00checked 1 hour ago | 131K33K max reply through OpenRouter | 22 tok/sthrough OpenRouter | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterConfirmed |
| Venice AIfp8Through OpenRouter | $0.45 / $3.50checked 1 hour ago | 128K16K max reply | 13 tok/s | No | No | Confirmed |
OpenRouter
Novita AIAcross the 5 listings we hold: 3 say they do not train on prompts (1 of them only through OpenRouter), 0 say they do and 2 do not say. 2 appear in the zero-retention registry we check (1 of them only through OpenRouter); the rest are unknown to us.
What each host's API supports
From the parameter list each endpoint publishes. Streaming is omitted: nothing we hold reports it, for any model.
| Provider | Tool calling | JSON output | Strict schema |
|---|---|---|---|
| OpenRouterOpenRouter's own listing | ✓ | ✓ | ✗ |
| DeepInfrafp8Direct | |||
| Alibaba CloudThrough OpenRouter | ✓ | ✓ | ✗ |
| Novita AIfp8Direct and through OpenRouter | ✓ | ✗ | ✗ |
| Venice AIfp8Through OpenRouter | ✓ | ✗ | ✗ |
Tool calling: 4 of 5 listings say yes, 1 publishes no parameter list. JSON output: 2 of 5 listings say yes, 2 say no, 1 publishes no parameter list. Strict schema: 0 of 5 listings say yes, 4 say no, 1 publishes no parameter list.
When we formed this view
Recent changes
What moved
first indexed by our pipelineEach date is the day we first saw the change, or the day the maker announced it.
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.
- 1 of 5 listings publishes no parameter list, so what its 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, and 1 answers only through OpenRouter, not for its own listing.
- We hold no cached-input rate for any of its listings.
- We hold no batch or off-peak rate for any of its listings.
- We hold a decode speed for it, but no prompt-processing (prefill) figure, so how long the input side of a job takes is unknown to us.
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/Qwen3-235B-A22B-Thinking-2507
- Architecture
- Mixture of experts
- Takes in, gives back
- Text in, text out
- Catalogue slug
- qwen-qwen3-235b-a22b-thinking-2507