Qwen3.5 397B A17B
Qwen · released Feb 16, 2026
The Qwen3.5 series 397B-A17B native vision-language model is built on a hybrid architecture that integrates a linear attention mechanism with a sparse mixture-of-experts model, achieving higher inference efficiency. It delivers...
Specification
- Context window
- 262K
- Max output
- 66K
- Knowledge cutoff
- Not stated
- Parameters
- 403.4B
- Licence
- apache-2.0
- Serving providers
- 11
- Moderated
- No
- Uptime
- 100.0%
Intelligence
34.3
80th percentile
Coding
48.2
Coding Index
Agentic
19.8
Agentic Index
Output speed
116 t/s
Median across providers
Latency
415ms
Time to first token
Cost per task
$0.21
Estimated
Benchmarks
Where the score comes from
The Intelligence Index is a composite. These are the underlying evaluations this model was actually measured on.
Evaluation scores
Percentage correct · higher is better
- τ²-bench (Telecom)95.6%
- GPQA Diamond89.3%
- IFBench78.8%
- AA-LCR (long context)72.7%
- SciCode42.0%
- Terminal-Bench Hard40.9%
- Humanity's Last Exam29.0%
An evaluation missing from this list was not run for this model — it is not a zero.
View as table
| Evaluation | Score |
|---|---|
| τ²-bench (Telecom) | 95.6% |
| GPQA Diamond | 89.3% |
| IFBench | 78.8% |
| AA-LCR (long context) | 72.7% |
| SciCode | 42.0% |
| Terminal-Bench Hard | 40.9% |
| Humanity's Last Exam | 29.0% |
Against its peers
Intelligence Index · this model highlighted, nearest peers in grey
- Qwen3.5-27BQwen34.6
- M2.5MiniMax34.5
- GLM 4.7Z.ai34.5
- Hy3 previewTencent34.4
- Qwen3.5 397B A17BQwen34.3
- LongCat 2.0Meituan33.9
- KAT-Coder-Pro V2KwaiPilot33.9
- Kimi K2 ThinkingMoonshot AI33.5
Peers are the models sitting closest on the Intelligence Index — the set you would realistically choose between.
View as table
| Model | Intelligence |
|---|---|
| Qwen3.5-27B | 34.6 |
| M2.5 | 34.5 |
| GLM 4.7 | 34.5 |
| Hy3 preview | 34.4 |
| Qwen3.5 397B A17B | 34.3 |
| LongCat 2.0 | 33.9 |
| KAT-Coder-Pro V2 | 33.9 |
| Kimi K2 Thinking | 33.5 |
Percentile among all indexed models
Pricing
What it costs to run
List prices per million tokens, plus what one representative task works out to.
List price
- Input / 1M tokens
- $0.39
- Output / 1M tokens
- $2.34
- Cached input / 1M
- Not offered
- Blended 3:1
- $0.877
One task, estimated
$0.21
- Input tokens
- 50,000
- Output tokens
- 80,000
- Profile
- Reasoning
Estimated from list pricing: 50K input tokens plus 80K output tokens for reasoning models (25K for non-reasoning).
Serving providers
Speed and latency figures are medians across these providers, so a widely-served model reports a blend rather than any single endpoint.
Arena
Head-to-head generation quality
Elo from pairwise judgements, broken out by the kind of thing the model was asked to build.
- Overall Elo
- 1,216
- Win rate
- 56.7%
- Strongest at
- 3D scenes
- Tournaments
- 960
Elo by category
Dot position on a 1,175–1,225 scale · Elo has no meaningful zero
- 1,216
- 1,213
- 1,202
- 1,198
- 1,188
- 1,181
Agent categories (full-stack apps, mobile apps) are only scored for models tested in agent harnesses.
View as table
| Category | Elo | Win rate |
|---|---|---|
| 3D scenes | 1216 | 56.7% |
| Websites | 1213 | 52.6% |
| Data visualisation | 1202 | 53.2% |
| UI components | 1198 | 51.4% |
| SVG | 1188 | 56.1% |
| Game development | 1181 | 50.1% |