Proton's Lumo 1.1: Powerful AI, Private by Design

In a world where many AI assistants seem to grow more data-hungry by the day, Proton is betting that you don’t have to trade privacy for performance. The Swiss privacy-focused company has rolled out Lumo 1.1, a major upgrade to its AI assistant that aims to deliver speed, intelligence, and accuracy without collecting and selling your data. If the early numbers hold, Lumo 1.1 isn’t just a marginal bump; it’s a deliberate leap toward a privacy-first mainstream alternative.

What’s new and why it matters

Think of Lumo 1.1 as a factory upgrade for a high-performance engine. Proton says the upgrade brings:

• A roughly 170% improvement in understanding the context of a user’s query, meaning your questions land closer to what you actually meant, with fewer follow-ups.
• A 40% uplift in generating and assisting with software code, which could appeal to developers and builders who want a private companion for scripting and debugging.
• An even more pronounced gain in planning and multi-step reasoning—reportedly over 200% better at choosing the right tools (like web search) and chaining actions to tackle complex prompts.

These aren’t cosmetic tweaks. Proton describes the changes as a fruits-of-the-technical-upgrade: updated underlying models, more processing power via GPUs, and a more capable tool-use framework that can handle projects, research tasks, and technical challenges with greater reliability.

For subscribers, the upgrade is reinforced by Proton’s subscription model. Lumo Plus users have access to the higher-performance tier, with the upgrades delivered on Proton’s own infrastructure rather than through outside partnerships.

The tech backbone: Eurostack and open-source compliance

At the core of Lumo 1.1 is a carefully curated set of open-source large language models. Proton discloses that Lumo 1.1 runs on a mix of models, including Mistral Small 3, Nemo, OLMO 2 32B, and OpenHands 32B. The key twist? These models are hosted entirely within Proton’s own European data centers, a framework the company calls its "Eurostack." The stated goal is transparency and data control: by keeping the models and the data path in-house, Proton can reduce the risk of broad surveillance and third-party mishandling of user information.

This approach is a clear departure from models that rely on external AI providers. Proton argues that running its own stack helps ensure user queries aren’t sent off to third parties for processing. The company even positions Eurostack as a privacy-first alternative to other private-cloud approaches—such as Apple’s Private Cloud Compute—by avoiding cross-border data routing and external collaborations.

In addition, Proton is pushing for verifiable security through openness. The company has announced that it will publish the source code for Lumo’s mobile applications, inviting the security community to audit privacy and security credentials. That kind of transparency can be rare in AI circles, and Proton says it’s about offering concrete ways for users to verify what happens to their data.

Privacy-forward by default

What makes Lumo stand out isn’t just the numbers; it’s the architecture that underpins it. Proton’s long-standing promise is a privacy-first design:

• Zero-logs by default: user conversations aren’t stored on Proton’s servers unless users opt in.
• Zero-access encryption for saved chats: even Proton can’t decrypt your stored conversations unless you provide the keys.
• No training data usage: user prompts aren’t used as training fodder for Lumo’s models, a common practice at larger AI firms.

The combination of these features means you can ask Lumo to plan a trip, pull current events, or brainstorm a new project without worrying that your private questions will show up in someone else’s search results.

What the upgrade means for users and the broader market

The Lumo 1.1 upgrade is positioned as more than a private tweak to a niche product. It’s a statement: you can have a capable AI companion that respects your privacy and still deliver on tasks that usually require data-sharing with large tech giants.

As AI tools mature, Proton’s model could push the market to rethink how”private” is defined in practice. If Lumo 1.1 continues to deliver on speed and accuracy gains—and if audits by the community back up Proton’s claims—the company might catalyze a broader shift toward in-house, user-controlled AI stacks.

Potential implications

• Increased pressure on big AI players to offer stronger privacy guarantees, or risk losing users who care about data sovereignty.
• A growing niche for subscription-funded, privacy-respecting AI tools that don’t rely on data monetization.
• Greater attention to open-source governance and independent verification of security claims.

Caveats and what to watch next

Proton’s numbers come from the company’s own testing and deployment data. Real-world performance can vary based on prompts, workload, and network conditions. Likewise, Eurostack’s effectiveness hinges on how well the selected models cooperate in diverse tasks and languages. Nonetheless, the push toward in-house, privacy-preserving AI is a trend that could gain traction as users demand more accountability and control over their digital footprints.

Final thoughts

Lumo 1.1 is more than a technical upgrade. It’s a strategic showcase of what AI can look like when performance isn’t funded by user data. The upgrade doesn’t just close the gap with some competitors—it reframes the conversation around privacy, control, and the future of AI as a consumer tool. If Proton can keep delivering on both speed and privacy, Lumo could become a visible alternative for people who want powerful AI without compromising what they value most: their privacy.

Key takeaways: Lumo 1.1 brings meaningful performance boosts, keeps data under user control, and demonstrates that privacy and power don’t have to be mutually exclusive.