Alibaba (Qwen) API Pricing
Alibaba sells the Qwen family through Model Studio, and it is the only provider tracked here whose cheapest tier goes below three cents per million input tokens. The catch is that where you deploy changes what you pay, by a lot.
| Model | Input / 1M | Output / 1M | Cached input / 1M | Batch | Notes |
|---|---|---|---|---|---|
| Qwen3.7 Flash | $0.03 | $0.13 | N/A | 50% | The cheapest model tracked anywhere on this site, on input price. |
| Qwen3.7 Max | $1.25 | $3.75 | N/A | 50% | Previous flagship, still listed and materially cheaper than Qwen3.8 Max. |
| Qwen3.8 Max | $2.00 | $6.00 | $0.25 | 50% | Alibaba's flagship, released August 3, 2026. Rates shown are the International (Singapore) endpoint; the Mainland China endpoint runs 60-70% cheaper. |
Alibaba's cheapest tracked model is Qwen3.7 Flash at $0.03 in / $0.13 out, and its most expensive is Qwen3.8 Max at $2.00 in / $6.00 out, a 55x spread across the lineup. Qwen3.8 Max ranks #41 of 60 models tracked across 17 providers.
Region changes the price more than the model does
Every rate on this page is for the International (Singapore) endpoint, because that is the endpoint most readers outside China can actually use. The Mainland China (Beijing) endpoint runs roughly 60 to 70 percent cheaper for the same models. That is a bigger swing than the gap between two adjacent Qwen tiers, so if you have a legitimate route to the Beijing endpoint, that decision will dominate your bill far more than which Qwen model you pick.
It also means any Qwen price you see quoted elsewhere is close to meaningless without knowing which endpoint it refers to. When comparing Qwen against a US provider, check that you are comparing the endpoint you would actually deploy against.
How the Qwen tiers stack up
Qwen3.8 Max is the current flagship, released on August 3, 2026, and priced in the same band as Grok 4.6 and Claude Sonnet 5 rather than at flagship-Western rates. Qwen3.7 Max, the previous flagship, is still listed and materially cheaper, which makes it a reasonable default for anything that was working fine before the 3.8 release.
Qwen3.7 Flash is the outlier. At three cents per million input tokens it is the cheapest model tracked anywhere on this site, cheap enough that for high-volume classification or routing the API cost stops being a line item worth optimising and the engineering time around it becomes the real expense.
Batch processing
Qwen models qualify for a 50 percent batch discount on both input and output, which stacks with an already low base rate. For anything latency-tolerant running at Qwen3.7 Flash rates, the effective cost approaches rounding error against the cost of the pipeline moving the data around.
Every Alibaba model on this site
| Model | Blended / 1M | Rank of 60 | Full pricing |
|---|---|---|---|
| Qwen3.7 Flash | $0.055 | #1 | Qwen3.7 Flash pricing page |
| Qwen3.7 Max | $1.88 | #33 | Qwen3.7 Max pricing page |
| Qwen3.8 Max | $3.00 | #41 | Qwen3.8 Max pricing page |
Alibaba pricing FAQ
Is Qwen cheaper than GPT or Claude?
Substantially, at every tier. Qwen3.8 Max sits at roughly the price of a mid-tier Western model despite being Alibaba's flagship, and Qwen3.7 Flash is cheaper than anything OpenAI, Anthropic, or Google sell. Whether it is a fair swap depends entirely on your task, and on whether deploying against a Chinese provider is acceptable for your data.
Why do Qwen prices differ so much between regions?
Alibaba prices the Mainland China (Beijing) endpoint 60 to 70 percent below the International (Singapore) endpoint. This site quotes Singapore rates. Both are official; they are just different products with different infrastructure and different markets behind them.
Does Qwen support batch discounts?
Yes, 50 percent off input and output for asynchronous batch jobs, the same headline discount Anthropic and Google offer. Use the batch API savings calculator to size it against your volume.