Proton Lumo Review (September 2026): Apertus Integration, In-Chat Charts, and the Updated Verdict

We rated Proton Lumo 4.0 in August: a genuinely private assistant that was good enough to work in, held back by a narrow audience. Two changes since then push it to 4.2. On September 17, Proton made Apertus 1.5, Switzerland's fully open sovereign model, a selectable option inside Lumo (proton.me). And the in-chat charting the August review flagged in a single line now has enough shape to judge for the one job it was built for. Neither turns Lumo into a frontier model. Both strengthen the case for the buyer who cannot legally paste client data into ChatGPT.

Here is the call before the detail. If your data can legally live on OpenAI or Anthropic servers, keep ChatGPT Plus or Claude Pro; the September updates don't change that. If it cannot, Lumo now pairs zero-access encryption with a Swiss-built, inspectable model and free private charting, and no mainstream competitor offers that combination.

What changed since August

Two things landed, and they aim at different parts of the same problem.

Apertus 1.5 arrived on September 17. Apertus is a fully open large language model from EPFL, ETH Zurich, and the Swiss National Supercomputing Centre, and Proton now lets you pick it from Lumo's model dropdown like any other option (proton.me). Its architecture, training data, and methods are public, and it was trained on openly available data with opt-out requests respected (ETH Zurich). For a team whose entire reason to be here is jurisdiction, a Swiss-origin model inside a Swiss product removes one more foreign dependency from the stack.

The second change is in-chat data visualization. Upload a spreadsheet, paste structured data, or just ask, and Lumo generates charts, key metrics, comparisons, and callouts inside the conversation, then keeps analyzing as you follow up (proton.me). It rolled out in early August (Releasebot), so the August review clocked it but didn't test the use case. The point that matters: the file and the chart both stay inside Proton's encrypted session, and the feature is free on every tier (proton.me).

What Lumo is

For anyone arriving without the August context: Lumo is Proton's private AI chat app, built in Switzerland, running open-source models on servers Proton controls in Europe under Swiss privacy law. Conversations use zero-access encryption, are never logged, and are kept out of model training, and the architecture is designed to support GDPR and HIPAA with ISO 27001 and SOC 2 certifications behind it (proton.me). Lumo 2.0, which shipped June 30, added a thinking mode, image generation, and persistent Projects, closing most of the everyday-work gap (Releasebot).

Pricing as of September 21, 2026

Nothing on the personal side moved since August. The business tier is on an intro discount worth flagging. All figures below are current as of September 21, 2026.

Plan	Price (USD)	What you get	Source
Proton Unlimited (bundles free Lumo)	$9.99/user/mo billed yearly ($119.88/yr), or $12.99/mo monthly	Full Proton stack: 500 GB, Mail, VPN, Drive, Pass, plus Lumo at the free tier's caps	proton.me/pricing
Lumo Plus	$12.99/mo, or $9.99/mo billed yearly	Lumo only: more Max-model usage, unlimited chats and Projects, unlimited Custom Lumos	proton.me/lumo/pricing
Lumo for Business (Professional)	$11.99/user/mo billed annually (20% intro off the standard $14.99)	Team seat management, admin controls, and org security on top of Lumo Plus	proton.me/business/lumo/pricing

The Business intro price is the one delta from August. At $11.99/user/mo it is well under ChatGPT Business, which sits at $20/user/mo billed annually with a two-seat minimum (Finout). Read the free tier honestly, though: Proton Unlimited includes Lumo at capped Max-model usage, capped messages, and one Project (proton.me/lumo/pricing). Light drafting fits inside that ceiling. Daily use hits the wall, and then you are paying for Plus.

What the Apertus integration actually changes

Set expectations first: Apertus is an option, not an upgrade. You select Apertus 1.5 from the model dropdown and chat as normal, with the same zero-access encryption over the session (proton.me). It does not raise Lumo's ceiling on hard reasoning; the open models here still sit a step behind GPT-5.x and Claude Opus on multi-step work, and adding an academic open model does not change that.

What it changes is the vendor-risk story for regulated buyers. Before, Lumo's privacy pitch rested on Swiss infrastructure running open-weight models like Qwen and GLM. Now a team can run a model whose training data and methods are public and Swiss in origin, inside a Swiss product, under Swiss law. For a procurement reviewer building a data-flow diagram, that is one fewer foreign jurisdiction to explain. Proton is blunt about the motive: "We don't want to spend the next decade relying on other countries for the AI we use every day," said Eamonn Maguire, Proton's Director of AI (proton.me). Lumo users can also send anonymized feedback to the university teams behind the model, which is mostly a win for Apertus rather than for the buyer (ETH Zurich).

In-chat data visualization: who benefits

The clearest winner is the compliance-bound analyst. Picture a security lead with a spreadsheet of quarterly incident counts by category. In ChatGPT, generating that chart means the data is processed on OpenAI's US infrastructure. OpenAI's business tiers do not train on that data by default (Finout), but the file still leaves your control and sits under US legal process, which is exactly the exposure a HIPAA or GDPR posture is meant to avoid. In Lumo, the same spreadsheet becomes a chart inside the encrypted session, and the file, the analysis, and the visual never leave Proton's infrastructure (proton.me).

Two prompts show the shape of what it does well:

Here is our Q3 incident log (attached CSV). Chart incident count by
category, then flag the two categories that grew fastest versus Q2 and
give me the percentage change for each.

Follow up in the same conversation to reshape the output:

Turn that into a summary table instead of a bar chart, sorted by
percentage change descending, and add a one-line callout naming the
single category I should brief the team on first.

Now the candid con. The charting is basic next to ChatGPT's data analysis. Lumo produces clean charts, metrics, and comparisons from reasonably structured input (proton.me), but ChatGPT's tool handles messier files, deeper transforms, and iterative cleanup with more range. If your work is heavy data wrangling and the data is not sensitive, ChatGPT is still the better instrument. Lumo's version wins on exactly one axis, and it is the axis this reader cares about: the data never leaves an encrypted, Swiss-jurisdiction session.

The verdict and the pick

Lumo earns 4.2, up from 4.0. The delta is two features that deepen the moat rather than widen the audience: a sovereign, inspectable model option and free private charting, both of which remove a specific step or dependency for the regulated buyer. What still caps it below the top is the audience size and the missing automation surface, not anything Proton shipped wrong.

Axis	Proton Lumo	ChatGPT Business (Team)
Data residency	Swiss/EU servers under Swiss law, zero-access encryption, outside US legal process (proton.me)	US infrastructure under US legal process; no training on business data by default (Finout)
Pricing	Lumo for Business $11.99/user/mo annual, intro off $14.99 (proton.me)	$20/user/mo annual, two-seat minimum (Finout)
In-chat charting	Yes, generated inside the encrypted session, free on all tiers; basic range (proton.me)	Yes, stronger and more flexible; data processed on OpenAI infrastructure (Finout)
Model options	Open models (Qwen, GLM) plus selectable Apertus 1.5 sovereign model (proton.me)	GPT-5.x family; no open-weight or sovereign option

The pick, by who you are:

Solo professional, no compliance requirement: keep ChatGPT Plus or Claude Pro. Apertus and charts are real, but the privacy premium buys you nothing your work needs, and the open models still trail on hard reasoning.

Regulated small team, three to fifteen people: Proton Unlimited at $9.99/mo billed yearly if you want the whole encrypted stack and only touch Lumo lightly, or Lumo for Business at $11.99/user/mo when you need the admin console and audit surface a HIPAA or GDPR review will ask for (proton.me/business/lumo/pricing). This is the buyer Lumo is built for, and the September updates make the case easier to sign off.

Enterprise security team: pilot it. Apertus plus Swiss jurisdiction narrows the gap with self-hosted inference far enough to be worth a controlled trial, and if you already run Proton VPN and Mail, the assistant slots into a stack your reviewers have already cleared. The missing API is the one thing that will decide it: if your workflows are automation-heavy, that gap is where the pilot ends.