Proton Lumo AI Review: Is the Private AI Worth Using?

Proton Lumo is the AI assistant built so that even the company running it cannot read your chats. It comes from Proton, the Swiss maker of Proton Mail and Proton VPN, and every saved conversation is protected by zero-access encryption under a strict no-logs policy. More than 10 million people have tried it. Since Lumo 2.0 it is no longer a slow privacy experiment. Proton says its top model, Lumo 2.0 Max, scores 51 on the Artificial Analysis Intelligence Index, up from 15 for Lumo 1.4.

This review covers what changed since Lumo 2.0 launched on June 30, 2026. You now pick between Lite, Max and Apertus 1.5 yourself, in Fast or Thinking mode, and there is image generation, long-term memory and Custom Lumos. It also checks how private Lumo really is, what Lumo Plus costs at $12.99 a month and how it compares with ChatGPT and Claude. Mac users get their own section, since there is still no native Mac app.

The Key Takeaways

• Zero-access encryption and no logs mean Proton cannot read your saved chats, and they are never used to train AI. The one exception is opt-in: if you choose to send feedback on an Apertus 1.5 answer, Proton anonymises it and shares it with the university team behind the model.
• You choose the model. Lumo 2.0 Lite is for everyday tasks, Lumo 2.0 Max for harder work, and Apertus 1.5 is a fully open Swiss model. Each runs in Fast or Thinking mode.
• Proton’s own launch figures put Lumo 2.0 Max at 51 and Lite at 34 on the Artificial Analysis Intelligence Index, against 15 for Lumo 1.4.
• The free tier works without an account. Lumo Plus is $12.99/month, or $9.99/month billed yearly, under ChatGPT Plus and Claude Pro at $20. Teams pay $11.99 per user/month for Lumo Professional, billed yearly.
• Lumo runs on the web, iPhone and Android, but there is no native Mac app. An independent audit also calls it the “least open” open AI assistant.

What Is Proton Lumo AI?

Proton Lumo is a privacy-first AI assistant from Proton, the Swiss company behind Proton Mail and Proton VPN. It runs open-source models on servers Proton controls, protected by zero-access encryption and a no-logs policy. Your conversations are never stored in readable form, shared, or used to train AI. Proton launched it in July 2025, and more than 10 million people have used it since. Proton AG is majority owned by the non-profit Proton Foundation, which is part of why the company sells privacy as the product, not a feature.

You can start chatting as a guest with no sign-up at all, and a guest chat is erased when the session ends. A free Proton account adds saved, encrypted chat history that syncs across devices, plus a Project to keep files and instructions together. Lumo handles the usual assistant work, writing, summarizing, brainstorming, research and light coding, and it can now read files from Proton Drive and turn a spreadsheet into a chart. The pitch is that you get that help without handing your data to an advertising company.

Lumo, GetLumo, and Luma AI Are Three Different Things

Search results mix up three unrelated products, so it is worth clearing this up before you download anything. Lumo by Proton is the privacy assistant covered here. GetLumo is a separate content-creation app for scripts and social posts. Luma AI is a video and 3D generation tool with no connection to Proton at all. If privacy is your goal, the one you want is published by Proton AG.

What Models Does Lumo Use?

Lumo gives you three models to pick from: Lumo 2.0 Lite, Lumo 2.0 Max and Apertus 1.5, as Proton’s guide to Lumo models and modes explains. Lite is built for quick everyday tasks, and Max is Lumo’s most powerful model for complex work. All three are open to every user, including the free tier. The difference between plans is the daily allowance for Max, which is much higher on Lumo Plus and Lumo for Business. Max is the default, and when you hit your daily Max limit Lumo switches you to Lite automatically.

Under those names sit open-weight models. Proton’s Lumo privacy page lists the models it currently uses as Qwen 3.5, GLM 5.3, Image-Turbo, FireRed-Image-Edit-1.1 and Apertus 1.5, all running only on servers Proton controls. Proton optimizes them and does not publish which one answers a given request. Earlier Lumo builds relied on a different lineup, including Mistral and OLMo 2, which is why many older reviews name models that are no longer current.

Fast and Thinking Modes

Every Lumo model runs in one of two modes. Fast answers as quickly as possible and suits lookups and simple questions. Thinking reasons through a request step by step before it answers, which is slower but better for complex problems. Both modes are available to all users on both Lite and Max. Proton’s own guide suggests Lite in Fast mode for quick questions, Max in Thinking mode for coding, and Max in Fast mode as the default when you are unsure.

Lumo Benchmark Scores

Proton published one benchmark in its Lumo 2.0 launch post. The Artificial Analysis Intelligence Index combines ten evaluations across agents, coding, scientific reasoning and general knowledge. On it, Proton reports 51 for Lumo 2.0 Max, 34 for Lumo 2.0 Lite and 15 for Lumo 1.4. That makes Max 240% higher and Lite 127% higher than the old version. These are Proton’s launch-day figures from June 2026, and Artificial Analysis has rescaled its index since, so compare them only with each other. Lumo’s ceiling still sits below the newest GPT and Claude Opus models, because open-weight models trail the top closed labs. For everyday writing and research that gap is small, but for the hardest coding or reasoning it shows.

Apertus 1.5, the Swiss Open Model

Apertus 1.5 joined Lumo on September 17, 2026. It is a fully open model built by researchers at ETH Zurich, EPFL and the Swiss National Supercomputing Centre. It is released under the Apache 2.0 licence, with its training data and code documented. Proton says it was trained on over 1,000 languages, including Romansh and Swiss German. Apertus chats get the same zero-access encryption as the rest of Lumo. The difference is that you can choose to send feedback on an Apertus answer, and Proton anonymises it and shares it with the research team. According to ETH Zurich’s announcement of the Lumo partnership, only users who give informed consent contribute.

What’s New in Lumo 2.0

The Lumo 2.0 release on June 30, 2026 is the version worth judging, and it is a real step up rather than a point update. Proton says it rebuilt the assistant on a new architecture, and CEO Andy Yen framed the new thinking mode as the headline capability. According to Proton’s Lumo 2.0 announcement covered by TechCrunch, the update responds to most queries up to 76% faster than the previous version.

The practical additions are the ones you notice day to day. Thinking mode handles harder problems and shows its reasoning as it works. Lumo can now analyze an uploaded image, generate one from a prompt or a rough sketch, and edit a photo you give it. Memory lets Lumo recall your preferences across sessions, and you decide what it keeps and what it forgets. The context window is twice as large, so long documents and long chats hold together. Web search is much stronger and cites its sources. Custom Lumos let you save a set of instructions as a dedicated assistant, such as a writer that always drafts in your tone.

Proton has kept shipping since the launch. A visual redesign followed on July 8, Lumo started turning spreadsheets and reports into data visualizations on August 3, and Apertus 1.5 arrived as a third model on September 17.

How Private Is Lumo, Really?

Lumo is private by design. Your prompts travel over TLS and are also encrypted so that only Lumo’s GPU servers can decrypt them. Once Lumo answers, the data on the server is erased, and Proton keeps no logs of what you asked or what Lumo replied. Saved chat history lives on your device and syncs to Proton with zero-access encryption, so even Proton staff cannot read it. Lumo runs on Proton’s European infrastructure under Swiss privacy law, and the app code is open source. You can read Proton’s own breakdown in its Lumo security model documentation.

For anyone handling sensitive material, legal notes, health questions, unpublished work, this is a different bargain from the one the mainstream assistants offer. ChatGPT and Gemini both retain conversations and, depending on your settings, train on them. Lumo removes that category of risk by design, which is the strongest reason to use it. The one exception is opt-in: feedback you choose to give on Apertus 1.5 answers is anonymised and shared with the Apertus researchers.

The “Least Open” Caveat You Should Know

Proton markets Lumo as open source, but that claim deserves a footnote. The European Open Source AI Index audited it and called it the “least open” open AI assistant in its database. Only the app code is public. The model weights, training data, system prompts, and the routing logic that decides which model answers you are all closed. That does not undermine the encryption, which is verifiable, but it does mean you are trusting Proton’s word on how the system behind the curtain works.

Lumo Pricing: Free, Plus, and Business

Lumo is cheaper than the mainstream assistants, and the free tier is usable rather than a demo. The catch is that the free limits on Max model use, messages, chat history and image generation arrive quickly if you lean on Lumo every day. Prices below are from Proton’s Lumo pricing page and its business pricing page, checked October 13, 2026.

At $12.99 a month, Lumo Plus undercuts both ChatGPT Plus and Claude Pro, which sit at $20. Billed yearly it drops to $9.99 a month. Businesses that already plan to move email and storage to Proton get Lumo inside Workspace Premium rather than as a separate bill.

Lumo vs ChatGPT and Claude

Lumo does not try to win on raw intelligence. It wins on trust. The table below sets Lumo Plus against the two mainstream leaders at their standard paid tiers, so you can see where each one fits. If you want a deeper platform breakdown, our guide on how the big three compare on Mac goes further.

The honest read is that Lumo is the best choice when privacy is the priority and a good enough answer beats the smartest possible answer. When you need the strongest coding or reasoning, or a native desktop experience, the mainstream tools still lead. For a wider view of the field, see our roundup of the best ChatGPT alternatives for Mac.

You Do Not Have to Pick One

A practical way to use Lumo is a split. Sensitive questions go to Lumo, and the hard coding problem or the long research task goes to a frontier model. You do not have to pay for two more subscriptions to get the second half. Fello AI puts ChatGPT, Claude, Gemini, Grok, DeepSeek and more in one native app for Mac, iPhone and iPad. When you need a stronger answer than Lumo gives, you ask the same question in a window that is already open. Keep Lumo for anything you would not paste into a mainstream assistant, since its zero-access encryption is the guarantee the others do not make.

Using Lumo on Mac and iPhone

Lumo has native apps for iPhone, iPad and Android, plus a full web app, but there is no native Mac app yet. On a Mac you run Lumo in the browser or add the web app to your Dock as a standalone window. That works well, but it lacks the system-wide shortcuts and menu-bar access that native clients offer. The mobile apps also have two things the web app lacks: voice input, and the ability to keep working on a request while you switch to another app. Your encrypted history syncs across all of them.

If a native, always-available assistant on your Mac matters to you, that gap is worth weighing. Many Apple users pair a privacy tool like Lumo with a multi-model app so they are not locked to one vendor. Our guide to the best AI apps for Mac covers the native options, and if you like the privacy angle you can also run other privacy-minded models on your Mac.

Is Lumo AI Any Good? Our Verdict

Lumo is very good at exactly what it sets out to do. For writing, summarizing, research and everyday questions, the answers are solid. The interface is calm and fast since the 2.0 update, and the privacy guarantee is the real thing rather than marketing. Being able to choose Max and Thinking mode yourself makes it far more useful than the version that launched in 2025. If you have ever hesitated before pasting something sensitive into ChatGPT, that hesitation disappears here, and that alone earns Lumo a place in your toolkit.

The limits are just as clear. The open-weight models will not match the frontier tools on the hardest coding and reasoning. The missing Mac app is a real gap for Apple users, and the “open source” label is thinner than it sounds. Our take is to use Lumo as your private default and keep a stronger model on hand for heavy lifting. You can see how the frontier models compare in our best AI models guide.

Conclusion

Proton Lumo is the clearest answer yet to a simple question: can you get a capable AI assistant without giving up your data? For most people the answer is now yes, and at $12.99 a month it costs less than the mainstream options. Start with the free tier and Lumo 2.0 Max to test the fit, and upgrade to Plus if the Max limit pinches. Keep a frontier model within reach for the jobs Lumo cannot handle.

FAQ