Qwen / Pricing
Qwen pricing:
API rates and usage costs.
Qwen API costs depend on the model, service region and size of each request. Use these clearly scoped rates and worked examples to build a realistic budget.
By China AI Chat Team · Sources checked · USD
Qwen chat and Qwen API have different billing
Start with the official Qwen assistant if you want a ready-to-use chat interface. Check the offer, features and limits shown in your account there. This guide’s token tables describe Alibaba Cloud Model Studio developer access, not a consumer chat subscription.
For a third-party host, use that provider’s invoice and price list. If you run a downloadable model yourself, budget for the machine, electricity or rented compute and ongoing operations; a model download is not a hosted inference service.
Scope of the tables: Alibaba Cloud Model Studio, Singapore region / International deployment, standard real-time token rates. They exclude temporary promotions, cache discounts and separate tool or infrastructure charges.
Selected Qwen API prices
USD per 1,000,000 tokens. Choose the row matching the input length of one request. This is a selected reference, not the complete model catalogue.
| Model ID | Input tokens per request | Input price | Output price |
|---|---|---|---|
qwen3.8-flash | Up to 1,000,000 | $0.15 | $0.47 |
qwen3.7-plus | Up to 256,000 | $0.40 | $1.60 |
qwen3.7-plus | Over 256,000, up to 1,000,000 | $1.20 | $4.80 |
qwen3.8-max | Up to 1,000,000 | $2.00 | $6.00 |
qwen3.7-flash | Up to 32,000 | $0.03 | $0.13 |
qwen3.7-flash | Over 32,000, up to 256,000 | $0.10 | $0.40 |
qwen3.7-flash | Over 256,000, up to 1,000,000 | $0.20 | $0.80 |
For the Plus and Max rows shown, the listed output rate applies in both thinking and non-thinking modes; billed thinking tokens count toward output. Alibaba applies the selected tier’s unit price to the whole request, rather than charging only the excess tokens at the higher rate.
Source: Alibaba Cloud model inference pricing, Singapore / International. Other regions and deployment scopes can have different rates. The Plus table uses list prices rather than the temporary discount displayed at review time.
Worked examples: estimate your own workload
1,000 short requests
Assume each request uses 2,000 input tokens and 500 billed output tokens, with no cache hits. Totals: 2 million input and 0.5 million output tokens. These are arithmetic examples, not measurements of model quality or task success.
| Model | Calculation | Estimated model charge |
|---|---|---|
| Qwen3.8 Flash | (2 × $0.15) + (0.5 × $0.47) | $0.535 |
| Qwen3.7 Plus | (2 × $0.40) + (0.5 × $1.60) | $1.60 |
| Qwen3.8 Max | (2 × $2) + (0.5 × $6) | $7.00 |
| Qwen3.7 Flash | (2 × $0.03) + (0.5 × $0.13) | $0.125 |
The tier is determined per request. The 2 million input tokens across this entire workload do not move these individual 2,000-token requests into a long-input tier.
One long request
A Qwen3.7 Plus request with 300,000 input tokens and 10,000 billed output tokens falls in the higher tier: (0.3 × $1.20) + (0.01 × $4.80) = $0.408 before other charges or discounts. Using the short-input rate would underestimate the bill.
For your own budget, multiply input and output separately, then add any tool, storage, hosting and applicable tax charges. A low price per token does not establish which model will finish your task successfully; test the same workload before committing to a provider.
Is there a free Qwen API allowance?
Alibaba documents a new-user allowance for eligible models in Singapore with International deployment. It is not a universal free allowance for every region or product. Current new-activation rules give a 90-day validity period; older accounts can follow earlier rules.
Check the exact model’s remaining allowance and expiry in your console. Input and output consume the same per-model quota. If your account is eligible for paid billing, usage beyond the allowance can become chargeable.
For a limited trial, check whether the model supports Free Quota Only and enable it before the allowance runs out if you want calls to stop instead of continuing as paid usage. Do not assume it is enabled by default.
Source: Alibaba Cloud free-quota rules. Separate coding/token plans and their dedicated keys follow their own billing arrangements.
How caching and batch affect the bill
Context caching
On supported models, cached input can cost less than ordinary input. Alibaba documents typical implicit-cache hits at 20% of the standard input rate. Explicit caching has different economics: creation is typically 125% of the normal input rate, while later hits are typically 10%.
Model support and cache rules matter. A repeated prompt does not guarantee an implicit hit. Use the actual cache counts and the model-specific rate when calculating savings; output is still billed separately. Read the official context-cache rules.
Batch processing
Eligible Batch File calls are priced at 50% of real-time inference, but support differs by model and region. A model appearing in the ordinary API table does not prove that the same model supports batch in Singapore. Check the supported-model list before applying that discount. Read the Batch File documentation.
Before you compare two Qwen prices
- Match the exact API model ID.
- Match the service region and deployment scope.
- Use the input tier for a single request.
- Separate input, output, cache and tool charges.
- Check whether the quote is a list price, promotion or different service plan.
This page reports sourced prices and our calculations. It does not claim a benchmark result or promise a fixed monthly bill.
Use the Qwen model guide to understand the available families, then follow the Qwen API guide to match your key, region and workspace.
Continue with the Qwen overview, compare Kimi pricing or DeepSeek pricing, and read our research approach. Report a pricing correction.