Alibaba Qwen's 2026 native vision-language Flash model. 1M context with strong multimodal agent execution at the lowest price point in the Qwen3.7 line.
Qwen3.7-Flash is the low-cost Flash tier of Alibaba's Qwen3.7 line. Alibaba Cloud Model Studio's visual-understanding guide recommends starting with qwen3.7-plus and switching to qwen3.7-flash to reduce costs once an application is stable, describing it as near-flagship performance with the same context length and feature set. The vendor lists it with a 1M-token context, image input up to 16 million pixels per image, video input up to 2 hours / 2 GB, and function calling, built-in tools, and structured output marked supported.
TheRouter lists the model as qwen/qwen3.7-flash on the OpenAI-compatible Chat Completions path. Runtime configuration comes from standard-models.yaml: 1,000,000-token context, 131,072 max completion tokens, text+image+video input, text output, and the parameters temperature, max_tokens, top_p, tools, tool_choice, response_format, and stop.
Qwen3.7-Flash is TheRouter's recommended successor for qwen/qwen3-vl-flash, which Alibaba retired in October 2026. It keeps a vision-capable, low-cost lane open while adding video input and a 1M-token context. Teams migrating from qwen3-vl-flash should regression-test OCR, grounding, and screenshot prompts before moving production traffic, because the two models are different checkpoints.
Alibaba Cloud Model Studio lists qwen3.7-flash with 1M context, thinking mode (toggled with enable_thinking), function calling, built-in tools such as web search and code execution, and structured output, plus image and video input.
TheRouter serves qwen/qwen3.7-flash on the OpenAI-compatible chat route with text+image+video input, text output, 1,000,000 context, 131,072 max output, and supported parameters temperature, max_tokens, top_p, tools, tool_choice, response_format, and stop. No reasoning parameter and no built-in-tool switch are listed; tools carries your own function definitions.
| Type | Rate |
|---|---|
| Prompts up to 32K tokens | |
| Input | $0.0324 / 1M tokens |
| Output | $0.1404 / 1M tokens |
| Cached input | $0.0324 / 1M tokens |
| Prompts over 32K tokens | |
| Input | $0.108 / 1M tokens |
| Output | $0.432 / 1M tokens |
| Cached input | $0.108 / 1M tokens |
Tiered pricing: when a request's prompt (input, cached tokens included) is over 32K tokens, every token in that request is billed at the long-context rates; exactly 32K is still the base rate. This model publishes no separate cached-input price: cached tokens are billed at the input rate of their group (the "Cached input" row shows that rate).
| Stable alias | qwen3.7-flash → qwen3.7-flash-2026-07-15alibabacloud.com ↗ | verified |
| Context window | 1,000,000 tokensalibabacloud.com ↗ | verified |
| Maximum completion | 131,072 tokens in standard-models.yaml | verified |
| Modalities | Text, image, and video input; text outputalibabacloud.com ↗ | verified |
| Image limits (vendor) | Up to 16M pixels per image; up to 256 images per requestalibabacloud.com ↗ | verified |
| Video limits (vendor) | Up to 2 hours / 2 GB per video; up to 64 videos per requestalibabacloud.com ↗ | verified |
| Thinking mode | Supported upstream (enable_thinking); TheRouter's parameter list for this model has no reasoning parameteralibabacloud.com ↗ | verified |
| Tooling | Function calling, built-in tools, and structured output listed as supported; Alibaba documents structured output for the Qwen3.7 series in non-thinking modealibabacloud.com ↗ | verified |
| Release date | Not publicly disclosed | unknown |
| Training cutoff | Not publicly disclosed | unknown |
| License / weights | Not publicly disclosed; no open-weight release found in reviewed Alibaba Cloud sources | unknown |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
Official benchmark table The reviewed Alibaba Cloud Model Studio pages position qwen3.7-flash by capabilities, limits, and context, but publish no benchmark table for it. Do not carry qwen3.7-plus or qwen3-vl-flash scores over to this model; run your own evals. | — | Not publicly disclosed | — |
Multimodal benchmark table No MMMU, OCRBench, ScreenSpot, or video-understanding score for qwen3.7-flash was found on the vendor pages reviewed. Alibaba's own wording is qualitative: near-flagship performance relative to qwen3.7-plus. | — | Not publicly disclosed | — |
Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.
curl https://api.therouter.ai/v1/chat/completions -H "Content-Type: application/json" -H "Authorization: Bearer $THE_ROUTER_API_KEY" -d '{
"model": "qwen/qwen3.7-flash",
"messages": [
{"role": "user", "content": "Summarize the key points from this input."}
]
}'Send images with the OpenAI-compatible content-array message format. This is the direct replacement call shape for former qwen/qwen3-vl-flash requests.
curl https://api.therouter.ai/v1/chat/completions \
-H "Authorization: Bearer $THEROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen/qwen3.7-flash",
"messages": [{
"role": "user",
"content": [
{"type": "text", "text": "Extract every visible error message from this screenshot."},
{"type": "image_url", "image_url": {"url": "https://therouter.ai/assets/vision-sample.png"}}
]
}],
"max_tokens": 1024
}'Model Studio's visual-understanding guide, updated October 11, 2026, recommends validating on qwen3.7-plus and then switching to qwen3.7-flash to cut costs, and lists qwen3.7-flash for general text extraction from images. The guide does not list qwen3-vl-flash.
Switch the model id to qwen/qwen3.7-flash; the request shape (OpenAI-compatible content array with image_url parts) stays the same. If you need a dedicated Qwen3-VL checkpoint instead, qwen/qwen3-vl-plus remains listed. Regression-test OCR and grounding prompts before moving production traffic.
Alibaba's guidance is to start with qwen3.7-plus and move to qwen3.7-flash once the application is stable and cost matters more than the remaining quality margin. Stay on Plus for requests that need more than 256 images, or where your evals show a quality gap that matters.
Alibaba documents thinking mode for Qwen3+ models via enable_thinking, but the supported parameter list for qwen/qwen3.7-flash does not include a reasoning control. Use only the parameters listed on the live model page.
No license or open-weight release for Qwen3.7-Flash was found in the Alibaba Cloud sources reviewed. Treat qwen/qwen3.7-flash as an API model unless Alibaba publishes a downloadable model card.
| source | URL | retrieved | |
|---|---|---|---|
| Stable alias | alibabacloud.com ↗ | 2026-10-17 | verified |
| Context window | alibabacloud.com ↗ | 2026-10-17 | verified |
| Maximum completion | — | — | verified |
| Modalities | alibabacloud.com ↗ | 2026-10-17 | verified |
| Image limits (vendor) | alibabacloud.com ↗ | 2026-10-17 | verified |
| Video limits (vendor) | alibabacloud.com ↗ | 2026-10-17 | verified |
| Thinking mode | alibabacloud.com ↗ | 2026-10-17 | verified |
| Tooling | alibabacloud.com ↗ | 2026-10-17 | verified |
| Release date | — | — | unknown |
| Training cutoff | — | — | unknown |
| License / weights | — | — | unknown |
| Official benchmark table | alibabacloud.com ↗ | 2026-10-17 | unknown |
| Multimodal benchmark table | alibabacloud.com ↗ | 2026-10-17 | unknown |
| Alibaba names qwen3.7-flash as the cost-reduced step after qwen3.7-plus for image and video understanding | alibabacloud.com ↗ | 2026-10-17 | verified |
| What should I use now that qwen/qwen3-vl-flash is retired? | alibabacloud.com ↗ | 2026-10-17 | to verify |
| When should I choose qwen3.7-flash instead of qwen3.7-plus? | alibabacloud.com ↗ | 2026-10-17 | to verify |
| Can I turn on thinking mode for qwen/qwen3.7-flash? | alibabacloud.com ↗ | 2026-10-17 | to verify |
| Is Qwen3.7-Flash open source? | alibabacloud.com ↗ | 2026-10-17 | to verify |