Models and providers

1. How your content travels

Beemm Vision is an orchestration platform. It does not train, host or operate artificial intelligence models: it routes your request to the model you select. Concretely, your prompt and any file you attach travel from Beemm Vision to a gateway, and from that gateway to the editor of the model.

This page identifies, for every model made available, the gateway used and the company that edits the model. It is the annex referred to in article 12 of the Terms of Sale and in section 7 of the Privacy Policy.

Last checked against the model registry: 24 October 2026.

2. The gateways

The choice of gateway is ours and may change for technical or tariff reasons, among the gateways listed for each model below. The choice of model editor is normally yours: it follows from the model you select, except in the cases listed in section 5.

3. Image and video models

4. Image editing operations

Upscaling, background removal, object erasing, inpainting and outpainting also call third-party models:

5. Language models

The assistant, the text and vision nodes of the workflows and the browser extension call language models through OpenRouter. Several of their editors are established outside the European Union, including in China (DeepSeek, Alibaba, Moonshot AI, Z.ai, MiniMax). The corresponding transfers are governed by Standard Contractual Clauses.

Cases where the editor is not the one you selected.

• Vision fallback. When a vision model refuses to describe an image, the same request (your prompt and the image) is sent again, by default, to Qwen 3 VL and then Qwen 3.7 Flash, edited by Alibaba (China).
• Automatic descriptions. The short description of a workflow is written by Gemini 2.5 Flash-Lite (Google).
• Withdrawn models. A model withdrawn by its editor is redirected to its successor for the workflows and requests that still name it: Grok 4 Fast to Grok 4.20, Claude 3.5 Sonnet to Claude Sonnet 5, and, for images and video, Veo 3.1 Fast to LTX 2.5 Fast, and the first Reve model and Ideogram V3 to Ideogram V4.

6. Browser extension analysis

The official Beemm Vision Chrome extension may observe image eligibility locally on ordinary HTTP(S) pages so a Prompt control can appear. No image data is sent on hover. After an explicit action and confirmation, the extension sends the selected image or screen region to a vision language model that you choose in the extension (default: Google Gemini 2.5 Flash). That request is routed through the OpenRouter gateway. A temporary copy may exist on our servers for the duration of the analysis and is deleted afterwards. Credit ledger entries and security logs remain.

7. Training on your content

Beemm does not train any model on your content and does not transfer your content to any third party for that purpose.

Whether a given editor uses the content it receives to train its own models is governed by that editor's own terms, over which we have no control. We configure our integrations so as to request, wherever the gateway or the editor offers the option, that transmitted content not be used for training. We cannot warrant that every editor abstains from doing so, and we will not claim otherwise.

Where this point is decisive for your use case, contact us before using a given model at [email protected] and we will tell you what we know of that editor's position, and what we do not.

8. Keeping this page accurate

This page is derived from the model registry actually deployed in production, and is updated whenever a model is added or withdrawn. The date of the last verification appears in section 1. If you notice a discrepancy between this page and what the interface offers, tell us at [email protected] — an out-of-date annex is a defect we want to know about.