The Qwen3.5 122B-A10B 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. In terms of...
Quick stats across all benchmark runs.
Models with similar or better quality but different tradeoffs
How this model performs across different benchmarks
Performance trends across all benchmark runs
Number of benchmark runs over time
Get started with this model using OpenRouter
Compare performance with other models from the same creator