Qwen3.5 397B A17B
Qwen · released Feb 16, 2026 · Qwen/Qwen3.5-397B-A17B
- Type
- Open weightsApache License 2.0
- Params
- 403B
- Context
- 262K
17B active per word · about 197K words of context
Our take
Written Sep 2, 2026Qwen 3.5 is a large mixture-of-experts model with 403.4 billion total parameters and 17 billion active per token, released under an Apache licence. It handles text, images and video in a single request of up to 262,144 tokens, and is available from 16 hosted providers.
Pick this for self-hosted frontier inference where a permissive licence matters, or for coding-heavy workloads where its measured score is strongest. Use it for long-context multimodal tasks, or when you want to trade price against speed across providers. Skip it if creative writing or web-development coding are your main needs, or if you need one provider that dominates both price and throughput.
The case for it
- Largest active-parameter count in the open-weights mixture-of-experts class we list: 403.4 billion total, 17 billion active.
- Coding is its strongest measured pillar, 49.9 points above its own overall chat score.
- Apache 2.0 licence allows commercial use, fine-tuning and redistribution at frontier scale.
- Wide provider choice with an 11× throughput spread and meaningful price variation.
The case against it
- Creative writing and web-development coding trail its own overall score by 36 and 43 points respectively.
- The cheapest tracked offer runs at only 6 tokens per second; faster options cost noticeably more.
- No single offer leads on both price and speed, so you must pick your trade-off.
How good is it?
EverydayGeneral questions and everyday reasoning
Arena Text (overall)57th of 168 · 1442
CodingWriting and fixing code on its own
Arena Coding58th of 168 · 1491
AgenticPlanning, calling tools, staying on task
Not yet scored on Arena Agent.
WritingDrafting and rewriting prose
Arena Creative Writing61st of 168 · 1405
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 model7 scoresEvery figure we hold, from 7 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. 120.6 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 1 hour ago — each listing carries its own date.
- per 1M tokens
- $0.39 in / $2.34 out
- Context served
- 262K
- Throughput
- ~71 tok/s
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| Alibaba CloudThrough OpenRouter | $0.39 / $2.34checked 1 hour ago | 262K66K max reply | 71 tok/s | No | Yesunknown period | Unknown |
| DeepInfrafp8Direct and through OpenRouter | $0.45 / $3.00checked 1 hour ago | 262K82K max reply through OpenRouter | 38 tok/sthrough OpenRouter | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterConfirmed |
| OpenRouterOpenRouter's own listing | $0.55 / $3.50checked 1 hour ago | 262K | not measured | Unknown | Unknown | Unknown |
| DigitalOcean GradientThrough OpenRouter | $0.55 / $3.50checked 1 hour ago | 131K118K max reply | 2 tok/s | No | No | Confirmed |
| AtlasCloudfp8Through OpenRouter | $0.55 / $3.50checked 1 hour ago | 262K66K max reply | 74 tok/s | No | Yesunknown period | Unknown |
| PhalaThrough OpenRouter | $0.55 / $3.50checked 1 hour ago | 262K236K max reply | 44 tok/s | No | No | Confirmed |
| StreamLakeThrough OpenRouter | $0.60 / $3.60checked 1 hour ago | 256K64K max reply | 131 tok/s | No | Yesunknown period | Unknown |
| Novita AIDirect and through OpenRouter | $0.60 / $3.60checked 1 hour ago | 262K66K max reply through OpenRouter | 63 tok/sthrough OpenRouter | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterConfirmed |
| GMICloudfp8Through OpenRouter | $0.60 / $3.60checked 1 hour ago | 262K236K max reply | 74 tok/s | No | Yesunknown period | Unknown |
| Parasailfp8Through OpenRouter | $0.50 / $3.60checked 1 hour ago | 262K236K max reply | 43 tok/s | No | No | Confirmed |
| Venice AIThrough OpenRouter | $0.75 / $4.50checked 1 hour ago | 128K33K max reply | 31 tok/s | No | No | Confirmed |
DeepInfra
OpenRouter
DigitalOcean Gradient
Novita AIAcross the 11 listings we hold: 10 say they do not train on prompts (2 of them only through OpenRouter), 0 say they do and 1 does not say. 6 appear in the zero-retention registry we check (2 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 |
|---|---|---|---|
| Alibaba CloudThrough OpenRouter | ✓ | ✓ | ✗ |
| DeepInfrafp8Direct and through OpenRouter | ✓ | ✓ | ✓ |
| OpenRouterOpenRouter's own listing | ✓ | ✓ | ✓ |
| DigitalOcean GradientThrough OpenRouter | ✓ | ✓ | ✓ |
| AtlasCloudfp8Through OpenRouter | ✓ | ✓ | ✓ |
| PhalaThrough OpenRouter | ✓ | ✗ | ✓ |
| StreamLakeThrough OpenRouter | ✗ | ✓ | ✓ |
| Novita AIDirect and through OpenRouter | ✓ | ✓ | ✗ |
| GMICloudfp8Through OpenRouter | ✓ | ✓ | ✗ |
| Parasailfp8Through OpenRouter | ✓ | ✓ | ✓ |
| Venice AIThrough OpenRouter | ✓ | ✓ | ✓ |
Tool calling: 10 of 11 listings say yes, 1 says no. JSON output: 10 of 11 listings say yes, 1 says no. Strict schema: 8 of 11 listings say yes, 3 say no.
Models people weigh against Qwen3.5 397B A17B
When we formed this view
Recent changes
What moved
input −21% ($0.3850 → $0.3025 per 1M tokens), output −21% ($2.45 → $1.93 per 1M tokens)What moved
cache read −80% ($0.225 → $0.045 per 1M tokens)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.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- 1 of 11 listings does not say whether it trains on prompts, and 2 answer only through OpenRouter, not for their own listing.
- 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/Qwen3.5-397B-A17B
- Architecture
- Mixture of experts
- Takes in, gives back
- Text, images and video in, text out
- Catalogue slug
- qwen-qwen3-5-397b-a17b