All models
Alibaba (Qwen)
Qwen3.5 397B A17B balancedruns locally
alibaba/qwen3-5-397b-a17b
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...
Input / 1M
$0.50
USD per million tokens
Output / 1M
$3.60
USD per million tokens
Est. cost / token
$0.000001275
$1.27 per 1M, 3:1 mix
Context window
262K
max output 236K
Major features
VisionOpen weightsCachingReasoningJSON schemaTools
Task suitability (heuristic 1–5, not a benchmark)
- coding4
- reasoning4
- math4
- summarization4
- extraction4
- classification4
- chat4
- creative4
- translation4
- vision4
- long context3
- agentic4
Run it locally
Will it run on my machine? →| Quantization | Quality | Memory needed (8K ctx) |
|---|---|---|
| Q2_K | poor | 184 GB |
| Q3_K_M | fair | 213 GB |
| Q4_K_Mrecommended | good | 265 GB |
| Q5_K_M | very good | 307 GB |
| Q6_K | excellent | 353 GB |
| Q8_0 | near-lossless | 456 GB |
| F16 | lossless | 853 GB |
Estimates. Mixture-of-experts: memory covers all parameters; only the active ones are used per token, so it runs faster than its size suggests.
Price history
| Recorded | Input | Output | Cached | Source |
|---|---|---|---|---|
| 2026-11-05 03:20 | $0.50 | $3.60 | $0.30 | openrouter |
A row is added whenever a sync detects a price change. Vendor pricing source
Pricing (USD per 1M tokens)
- Input
- $0.50
- Output
- $3.60
- Cached input
- $0.30
- Blended (3:1)
- $1.27
Details
- Company
- Alibaba (Qwen)
- Version
- 3.5
- Family
- Qwen
- Released
- 2026-02-16
- Parameters
- 403.4B (17B active)
- Max output
- 236K
- Open weights
- Yes
- Inputs
- image, text, video
- Last synced
- 25 min ago