Proton Lumo 2.0: The Private AI That Competes with ChatGPT Without Using Your Data
The update doubles performance, adds multimodality and encrypted memory, while maintaining the promise of zero training with user data.
July 4, 2026 · 4 min read

TL;DR: Proton Lumo 2.0 offers performance comparable to ChatGPT and Claude, with the key difference that it never trains on user data. It incorporates multimodality, controlled memory, and zero-access encryption.
What happened?
Proton, the Swiss company known for ProtonMail and ProtonVPN, has launched Lumo 2.0, the second generation of its artificial intelligence assistant. The announcement, covered by ZDNet, WWWhat's new, TechCrunch, 9to5Mac and Engadget, comes a year after the debut of Lumo 1.0 in July 2025. The new version incorporates multimodality (image analysis and generation), reasoning mode, user-controlled memory, encrypted projects, and a doubled context window.
According to data from the Artificial Analysis Intelligence Index —an independent benchmark combining ten evaluations of reasoning, agents, programming, and general knowledge— Lumo 2.0 Lite scores 127% higher than Lumo 1.4, while Lumo 2.0 Max achieves 240% higher. Proton itself states that “in many use cases, users can no longer perceive a qualitative difference between Lumo 2.0 Max and the latest models from OpenAI and Anthropic.” It is worth noting that these figures are provided by the manufacturer, although the underlying benchmark is external and auditable.
Why is it important?
Proton's value proposition has always been privacy. Unlike ChatGPT, Claude, or Gemini, Lumo never trains its models with user data. All processing is done under zero-access encryption: not even Proton can see the conversations. This is especially relevant in a context where the European Union strengthens data protection with GDPR and where more companies seek alternatives that comply with strict regulations, such as the EU AI Act or HIPAA in the United States for the healthcare sector.
The leap in quality is key. Until now, privacy often came with inferior performance. Lumo 2.0 breaks that barrier by offering capabilities comparable to market leaders. If the benchmarks hold, Proton could attract users who previously sacrificed privacy for functionality. In 2025, when Lumo 1.0 launched, analysts noted a significant performance gap compared to GPT-4o and Claude 3.5 Sonnet. Now, with Lumo 2.0, that gap has narrowed dramatically, potentially changing competitive dynamics in the AI assistant market.
Key technical innovations
- True multimodality: upload images for analysis or editing, and generate images from text, all encrypted. This includes the ability to edit images with text instructions, similar to what DALL-E 3 or Gemini offer, but with the difference that no data leaves Proton's encrypted environment.
- Reasoning mode: for complex questions, the model spends more compute time thinking before answering. This feature, similar to OpenAI's o1 or Claude 3.5 Sonnet, allows tackling mathematical, logical, or programming problems with greater accuracy.
- User-controlled memory: the user decides what Lumo remembers, forgets, and never learns. Unlike ChatGPT, which stores conversations by default, Proton allows selectively deleting memories or disabling memory entirely.
- Projects: encrypted workspaces that group conversations, files, and instructions by topic. This functionality is similar to Claude's projects, but with the advantage of zero-access encryption.
- Improved web search: real-time results with privacy respect. Proton claims the search does not log IP or history, unlike traditional search engines.
What consequences will it have?
In the short term, Lumo 2.0 pressures OpenAI, Anthropic, and Google to offer stronger privacy guarantees. In the long term, it could accelerate AI adoption in regulated sectors like healthcare, finance, and public administration, where regulatory compliance is critical. For example, a hospital handling patient data under HIPAA might prefer Lumo 2.0 over ChatGPT, which has been criticized for leaking user data in the past. Similarly, law firms handling confidential information could see Lumo as a viable alternative.
However, caution is warranted. The benchmarks are provided by the manufacturer itself, albeit based on independent indices. Real-world user experience may differ. Additionally, Proton does not disclose which underlying models Lumo 2.0 uses, making external verification difficult. In the past, other companies have exaggerated their proprietary benchmarks. For instance, in 2023, some benchmarks of open-source models turned out to be inflated when compared to specific versions of GPT-3.5.
“Privacy should not be a luxury, but a right. With Lumo 2.0, we demonstrate that you can have top-tier AI without compromising user data,” said a Proton spokesperson.
What readers should know
Lumo 2.0 has been available since July 2026 for all Proton subscribers (free and paid plans). The free version has limitations on the number of queries and access to advanced features like reasoning mode. For those prioritizing privacy, this is the most solid alternative to ChatGPT on the market today.
However, Proton's ecosystem is limited: it lacks the massive third-party integration that ChatGPT has, nor does it have Anthropic's ecosystem. Advanced AI users may miss plugins or connectors with productivity tools. Additionally, Proton does not offer a public API for developers, limiting its use in custom enterprise applications. Compared to the launch of ChatGPT Enterprise in 2023, which did offer API and administrative controls, Lumo 2.0 seems more oriented toward the end consumer than large corporate deployments.