Model
Qwen2.5-VL 72B (vision)
EU sovereignTemporarily pausedQwen2.5-VL 72B (vision): temporarily paused on Frontière AI, back unchanged.
€0.1529
Input — per 100k tokens
€0.1529
Output — per 100k tokens
About this model
Qwen2.5-VL-72B-Instruct is the 72-billion-parameter instruction-tuned model of Qwen's Qwen2.5-VL vision-language series, released in January 2025. It takes text, images and video and answers in text. According to its model card, it reads text, charts and layouts in images, follows videos longer than an hour, locates objects as bounding boxes in JSON, and extracts structured data from invoices, forms and tables; it reports 96.4 on DocVQA. Its licence is the Qwen License Agreement, not Apache 2.0: commercial products with over 100 million monthly active users must request a licence. It suits document and image extraction, such as invoices, forms and charts. Qwen3.6 27B is a newer multimodal Qwen model in this catalogue.
Sources, checked on 23 September 2026: huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct ↗ · Qwen License Agreement (LICENSE) ↗
Specifications
- API model id
- qwen2.5-vl-72b
- Availability
- Temporarily paused
- Hosting jurisdiction
- EU sovereign
- Served via
- OVHcloud
- Publisher
- Qwen
- Parameters
- 73.4B
- Licence
- Qwen License Agreement
- Context window
- 32k
- Modality
- Vision-language
Vision model — its provider rejects tool calls with a 400, so it cannot back an agent. Use it for image understanding, not for function calling.
Open weights & AI Act facts
Factual characteristics of this model's weights — stated, not certified. Under the EU AI Act, whether a deployment is compliant depends on the model, your role and your use case.
- Open weights
- Weights public — use governed by the licence above
- AI Act class
- GPAI — general-purpose AI
- Systemic risk
- Not assessed by us — training compute not published, or too close to the 10²⁵ FLOP threshold to call
- Data retention
- None at Frontière AI; OVHcloud states zero data retention (billing data only)
What its label means
Paused: OVHcloud, the only European provider serving this model, is currently unavailable to us. Use pixtral-12b (vision, served by Scaleway) or qwen3.6-35b meanwhile.
How we draw the line: GDPR vs. the CLOUD Act →All Qwen models and their jurisdiction →Pricing
Paused: nothing can be called, so nothing is billed right now. The price above is the one that applies, unchanged, when it comes back — margin and VAT included.
Compare every Qwen model's API price →FAQ
Is Qwen2.5-VL 72B (vision) GDPR-compliant on Frontière AI?
Qwen2.5-VL 72B (vision) is temporarily paused. Paused: OVHcloud, the only European provider serving this model, is currently unavailable to us. Use pixtral-12b (vision, served by Scaleway) or qwen3.6-35b meanwhile. It will carry the EU sovereign label again when it comes back.
Under which licence is Qwen2.5-VL 72B (vision) released?
Qwen License Agreement — this is not a permissive Apache- or MIT-style licence. Worth flagging if you run a legal review of licences before deployment.
Is there any behaviour I should know about?
Vision model — its provider rejects tool calls with a 400, so it cannot back an agent. Use it for image understanding, not for function calling.
Articles about this model
Other models in the catalog
Call it in minutes
Create an account, top up €10, change two lines.