Qwen3 Max
Qwen · released Sep 23, 2025
- Type
- Closed
- Input
- None held
- Output
- None held
- Cached
- None held
We don't hold a list price for this model yet · hosted only — we have no record of published weights
Our take
Written Sep 17, 2026Qwen3 Max is a hosted-only text model: we list no download for it, so using it means choosing a host. Its measured strengths sit in mathematics and reasoning, and its weakest measured area is agentic coding.
Reach for it on mathematical and reasoning-heavy work, where it scores 91.31% on LiveBench mathematics and 88.21% on reasoning, or when long documents need not be split up first. Four companies serve it, so you can move the workload if a host's rate stops suiting you. Skip it if you need to run the model on your own hardware, or if your work is agentic coding inside a harness.
The case for it
- Measured strength on mathematics and reasoning: 91.31% on LiveBench mathematics and 88.21% on reasoning, so competition-style problems are where it earns its place.
- Language comprehension and data analysis hold up at 79.69% and 78.41%, which suits reading and table work rather than only sums.
- The request capacity takes long documents without splitting them first, though reliable recall across all of it is unverified in our data.
The case against it
- Agentic coding is its weakest measured area at 64.65% on LiveBench, run inside an agent harness, so tool-use chains are the wrong first job for it.
- We list no download for it, so running it yourself is not a route we can point you to and a host is the only option.
- The licence is not stated in our data, so commercial use and redistribution permissions are unverified and need checking before you build on it.
How good is it?
EverydayGeneral questions and everyday reasoning
Arena Text (overall)66th of 168 · 1435
CodingWriting and fixing code on its own
Arena Coding68th of 168 · 1480
AgenticPlanning, calling tools, staying on task
Not yet scored on Arena Agent. It is on LiveBench Agentic Coding, in 6th of 58 with 64.65.
WritingDrafting and rewriting prose
Arena Creative Writing75th of 168 · 1393
Boards this model appears on that none of the ratings above are built on.
Every published score for this model14 scoresEvery figure we hold, from 14 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Where to rent it
Prices checked 4 hours ago — each listing carries its own date.
- per 1M tokens
- $0.78 in / $3.90 out
- Context served
- 262K
- Throughput
- ~50 tok/s
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| OpenRouterOpenRouter's own listing | $0.78 / $3.90checked 4 hours ago | 262K | not measured | Unknown | Unknown | Unknown |
| Alibaba CloudThrough OpenRouter | $0.78 / $3.90checked 4 hours ago | 262K66K max reply | 50 tok/s | No | Yesunknown period | Unknown |
| DeepInfraDirect | $1.20 / $6.00checked 4 hours ago | 256K | not measured | Unknown | Unknown | Unknown |
| Novita AIDirect | $2.11 / $8.45checked 4 hours ago | 262K | not measured | Unknown | Unknown | Unknown |
OpenRouter
DeepInfra
Novita AIAcross the 4 listings we hold: 1 says it does not train on prompts, 0 say they do and 3 do not say. 0 appear in the zero-retention registry we check; 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 | ✓ | ✓ | ✓ |
| Alibaba CloudThrough OpenRouter | ✓ | ✓ | ✓ |
| DeepInfraDirect | |||
| Novita AIDirect |
Tool calling: 2 of 4 listings say yes, 2 publish no parameter list. JSON output: 2 of 4 listings say yes, 2 publish no parameter list. Strict schema: 2 of 4 listings say yes, 2 publish no parameter list.
Models people weigh against Qwen3 Max
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
- 2 of 4 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 4 listings do not say whether they train on prompts.
- We don't hold a list price for this model yet — the gap is ours, not the lab's.
- We hold no batch or off-peak rate for any of its listings.
Licence and identifiers
What the licence allowsWe hold no licence record for this model. Inside are the identifiers you need to pull it — its Hugging Face repo where we have one, our slug and a machine-readable card.
Licence
We hold no licence record for this model, and no record of published weights either — so we can neither summarise its terms nor point you at the weights.
Identifiers
- Takes in, gives back
- Text in, text out
- Catalogue slug
- qwen-qwen3-max