The Qwen3.5 native vision-language Flash models are built on a hybrid architecture that integrates a linear attention mechanism with a sparse mixture-of-experts model, achieving higher inference efficiency. Compared to the...
Rank
#73
Up 1% week over week
Tokens
60.5B
2026-08-14
Requests
25.4M
25,430,589
Tool calls
-
Input / M
$0.07
Output / M
$0.26
Cache / M
Free
Context
1M
1,000,000 tokens