Model positioning and release date
Qwen-Flash is a text generation and reasoning model in the Qwen3 series. The official international site lists qwen-flash and the snapshot version qwen-flash-2025-07-28 in its update of August 1, 2025; July 28, 2025 in the snapshot name is the snapshot date. The official model page states that the current qwen-flash has capabilities equivalent to that snapshot. (alibabacloud.com)
Capabilities and uses
- Supports text input and text output, combining thinking and non-thinking modes that can be switched during a conversation; the official source states that its complex reasoning, instruction following, and text comprehension capabilities have improved.
- Suitable for text generation, text comprehension, question answering, and tasks requiring reasoning. Specific feature support varies by region, deployment scope, and model version.
- The official model page lists a context length of 1,000,000 tokens, a maximum input of 997,952 tokens, and a maximum output of 32,768 tokens for the snapshot version. (help.aliyun.com)
Limitations and availability
The official domestic model page states that the snapshot version in the Beijing region supports function calling, structured output, web search, and prefix completion, but does not support context caching, batch inference, or model tuning. The snapshot versions in the Singapore, Frankfurt, and Virginia regions listed on that page do not support function calling or web search, nor do they support the caching, batch inference, and tuning features mentioned above. Do not directly apply the dynamic alias's feature discounts to the snapshot. (help.aliyun.com)
An international-site retirement notice lists qwen-flash-2025-07-28-us and qwen-flash-us, which have the -us suffix, among models scheduled for retirement on October 10, 2026, and notes that the actual change time will prevail. This notice does not imply that models in other regions or models without the suffix have all been retired. Confirm availability in the console for the corresponding region before use. (alibabacloud.com)
Official pricing
The following are the real-time inference unit prices listed on official pages. The tier is determined by the number of input tokens per request; input and output are billed separately, both per million tokens. Prices and availability may change. Date checked: November 18, 2026.
Domestic site: Chinese yuan (CNY)
Alibaba Cloud Model Studio China (Beijing), applicable to qwen-flash and qwen-flash-2025-07-28, with the same prices for thinking and non-thinking modes:
| Input tokens per request | Input price | Output price |
|---|---|---|
| 0 < input ≤ 128K | ¥0.15 | ¥1.5 |
| 128K < input ≤ 256K | ¥0.6 | ¥6 |
| 256K < input ≤ 1M | ¥1.2 | ¥12 |
Source: Alibaba Cloud Model Studio domestic model pricing. (help.aliyun.com)
International site: US dollars (USD)
Alibaba Cloud Model Studio, Singapore region, international deployment scope, applicable to qwen-flash and qwen-flash-2025-07-28:
| Input tokens per request | Input price | Output price |
|---|---|---|
| 0 < input ≤ 256K | $0.05 | $0.4 |
| 256K < input ≤ 1M | $0.25 | $2 |
The international site also lists unit prices for the US Virginia region, global deployment scope: for input up to 128K, $0.022 for input and $0.216 for output; from 128K to 256K, $0.087 for input and $0.861 for output; from 256K to 1M, $0.173 for input and $1.721 for output, also per million tokens. Different regions and deployment scopes are not the same plan and cannot be directly converted into one another. (docs.modelstudio.console.alibabacloud.com)
Source: Alibaba Cloud Model Studio international model pricing. The prices above do not include separately applicable caching, batch processing, or limited-time offers; features and discounts must be checked against the specific version and region. (help.aliyun.com)