The mannequin retains the two.4-trillion-parameter mixture-of-experts spine of its predecessor, Qwen3.8-Max, however prompts about 71 billion parameters per token as a substitute of about 95 billion, the corporate stated. It helps a 1-million-token context window.
The corporate attributed the financial savings to decrease token use and new record costs of $1.50 per million enter tokens and $4.50 per million output tokens, down from $2 and $6. Anthropic lists Claude Opus 5.5 at $20 per million output tokens.
Sanchit Vir Gogia, chief analyst at Greyhound Analysis, stated enterprises ought to deal with the determine as a vendor benchmark relatively than an assured discount in their very own AI spending. The related measure, he stated, is the price of an accepted consequence, together with retries, reasoning tokens, and worker time spent checking outputs.


