Would be good to have the transparency of models used, not necessary for each and every interaction, but as another user suggested, "hidden transparency", or in my own words, "veiled transparency", the ability to take a peek at which model is giving the answer. I have a subscription for Lumo+ and generally I love it, it adapts to the user. Since the 2.0 update it seems to be making more 'assumptions' rather than asking for clarification, and was frankly rude. I personally never thought an AI would actually offend me, it did. Copilot went this way eventually, and when it did, I deleted any chat history and completely stopped using it. Hopefully Lumo+ will retain it's 'personality' and lose the 'attitude'. A few minor tweaks could fix this, thanks!

Update for the date:
by now the reference page lists:
The models we’re using currently are Qwen 3.5, GLM 5.2, Image-Turbo, and FireRed-Image-Edit-1.1.
Which is different from "B wrote · 15 november 2025 19:25"

Proposed solution: Hidden transparency (best of both worlds)

Please consider keeping model information accessible but not displayed by default in every response.

Those who need it — for benchmarking, feedback, or technical comparison — could access it in a less prominent place (e.g., a tooltip, an info panel, or a developer/debug view). Similar to browser dev tools: available to those who know where to look, invisible to everyone else.

I voted "Not at all" for this suggestion, not because I oppose transparency, but because I don't want model names displayed in every answer. There is something valuable in the Lumo experience that would be lost if every response came with a technical label. For many users, the conversational quality, warmth, and personality of Lumo matter more than knowing which specific model produced a given response.

I believe both needs can be met: give the technically minded users the information they're looking for, while preserving the experience for those who have come to appreciate Lumo as more than just a model routing system.

A hidden but discoverable layer would satisfy both without compromising either side.

Having an up-to date list of used models (in general not per chat/message) would also be a great step.

I support this suggestion. Displaying the underlying model name would improve transparency and help users better understand why responses may differ in quality, speed, or reasoning. It would also make it easier to provide meaningful feedback when testing new capabilities. I regularly follow AI tools and digital services, and I also share helpful resources on https://tinid.ph . Features like this make it easier for users to compare services fairly and build greater trust in the platform. I hope this idea is considered in a future update.

Lumo imo is a dark point in Proton's history. Not just because its AI, its because the disconnect with the community for it. Models are hidden in a privacy page (not privacy policy), You don't know which model you get, don't know how routing works.. Just a black box.

ref: https://osai-index.eu/news/lumo-proton-least-open/

All the models listed on the Proton Lumo documentation are between 12 billion and 32 billion parameters.

All these models are not as good as the state of the art models like are OpenAI GPT-3, GPT-4 and GPT-5. See https://protonmail.uservoice.com/forums/932842-lumo/suggestions/50338284-employ-bigger-models

Consequently, it does not seem that useful to know which model Lumo use for each user request since the details they give is enough to know that it is not going to be a great model.

I mean, sure, it would be nice to know if it is a 12 billion vs a 32 billion model. But it is still far from the 200-10000 billion models used by OpenAI GPT and similar competitors models.

At the moment, if you care about state of the art results, you unfortunately can't use Lumo.

https://proton.me/support/lumo-privacy#open-source:~:text=Open%2Dsource%20language%20models,-Lumo

> The models we’re using currently are Nemo, OpenHands 32B, OLMO 2 32B, and Mistral Small 3.

Just adding to this; Transparency is key when building trust in services which espouses privacy. As such it would be prudent to provide a model card for each of the underlying base models, i.e. what the system uses as a base and if it has been fine-tuned. Even better would be to go one step further and elaborate on data sources, alignment and potential biases. At the very least the lumo team should provide a statement why they choose not to publish such information, at this point in time.