Qwen3.5-122B-A10B is a native multi-modal large language model launched by Tongyi Qianwen team, with a total parameter amount of 122B and an activation parameter amount of only 10B. The model uses an efficient hybrid architecture that combines a gated delta network with a sparse mixture of experts (MoE), natively supports 256K context length and is scalable to approximately 1 million tokens. The model achieves unified basic visual language capabilities through early fusion training...
Provider: Qwen
Context window: 256K
Pricing: $0.24 input / $1.92 output per 1M tokens
Capabilities: serverless, json_mode, function_calling, structured_output
