Qwen API pricing for every model, newest first

Alibaba sets prices per deployment region in Model Studio, so the rate where you deploy can differ from the table. Rows are per million tokens.

Some Qwen models charge by prompt size. Qwen3.7 Plus bills a prompt past 256,000 tokens at $0.96 input and $3.84 output per million, so a long document and a short chat don't cost the same per token. Cached input on Qwen3.8 Max Prime costs $0.50 per million, 88% off the normal rate.

Work that can wait goes through the Batch API. On Qwen3.8 Max (0902) that's $1.15 input and $3.45 output per million, 43% off input.

Alibaba also publishes open weights for many Qwen models, usually the ones named by size, like 27B or 35B-A3B. That's why so many rows show a host, and why you can run those yourself. The largest window belongs to Qwen3.8 2.4T A95B, at 1.05M tokens, against 1M on Qwen3.8 Max Prime.

Start with Qwen3.8 Max Prime and run Qwen3.5-Flash, the lowest cost model Alibaba sells for the example request, on the same prompts. Use it wherever it passes your checks. Save Qwen3.8 Max (0902) for the jobs that need its size.

Check the rate in your deployment region before you set a budget, and keep long prompts under the long prompt line where you can.