Back to Explore
Qwen: Qwen3.5 397B A17B
Alibaba·Released 2026-02-16
Open Source397B256K ctxApache 2.0Multimodal
About
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...
Related Models
Pricing
Input
$0.39
per 1M tokens
Output
$2.45
per 1M tokens
Blended
$0.90
per 1M tokens
Cheaper than 37% of models. Median price is $0.54/1M tokens.
Cost Calculator
Tokens per day1M
100K100M
Daily
$0.90
Monthly
$27.04
vs. Similar Models
MiniMax: MiniMax M2.5Q:0.0
$0.21-77%
Claude 4.1 Opus (Reasoning)Q:0.0
$30.00+3229%
GPT-5 (medium)Q:0.0
$3.44+281%
Qwen: Qwen3.5-27BQ:+0.1
$0.54-40%
Performance
52
tokens/sec
Faster than 18% of models
1.68
seconds
Faster than 30% of models
63.15
seconds
Faster than 4% of models
Market Median
94 tok/s
45% slower
Median TTFT
1.11s
51% slower
Throughput/Dollar
58
tok/s per $/1M
Speed Comparison
Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)
52 tok/s-0%
Claude 4.5 Sonnet (Reasoning)
52 tok/s-0%
Ling-flash-2.0
52 tok/s+0%
Context Window
256K
tokens
Larger than 58% of models
Max Output
64K
tokens
25% of context
Benchmarks
MMLU-ProNot evaluated
GPQA Diamond
89.3%
HLE
27.3%
LiveCodeBenchNot evaluated
SciCode
42.0%
TerminalBench Hard
40.9%
MATH-500Not evaluated
AIMENot evaluated
AIME 2025Not evaluated
IFBench
78.8%
Long Context Recall
65.7%
Tau2
95.6%
Market AverageTop Score
Open Source
apache-2.0397BGGUF / GPTQ / AWQ
Downloads
590.0K
Likes
1.5K
VRAM (FP16)
Multi-GPU
GPU
8x A100 / H100