Lumo

Lumo is Proton’s general-purpose AI assistant, now on the Lumo 2.0 generation. The current product has moved well beyond its privacy-focused 2025 launch: it offers Fast and Thinking modes, web search with citations, file and image understanding, image generation and editing, longer-term memory, Projects, Custom Lumos and data-visualization workflows.¹²³ It is available on the web and through iOS and Android apps, and basic use can begin without creating a Proton account.⁴³⁵

That breadth matters because Lumo’s original differentiation was mostly privacy. Lumo 2.0 is attempting to compete as a full everyday assistant while preserving that privacy architecture. TechCrunch’s current 2.0 coverage independently confirms the expansion into image recognition and generation, Projects-based memory and a dedicated Thinking mode.⁶ TechRadar likewise describes Lumo 2.0 as a substantial capability upgrade rather than a minor interface refresh.⁷

Lumo should not be evaluated as one fixed foundation model. Proton’s current privacy documentation lists a routed open-source stack that includes Qwen 3.5, GLM 5.2, Image-Turbo and FireRed-Image-Edit-1.1, with Proton optimizing and running models on infrastructure it controls.⁸ Different tasks can therefore reach different models, and that model mix can change. A benchmark score for Qwen or GLM is not automatically a Lumo product score.

Proton publishes its own Lumo 2.0 performance claims. It reports that the Lite configuration improved 127% and Max 240% over Lumo 1.4 on an Artificial Analysis Intelligence Index-based comparison and says everyday questions can complete up to 76% faster.² Those are useful indicators of the size of Proton’s claimed generational improvement, but they are vendor-run comparisons rather than a direct independent Artificial Analysis evaluation of the complete Lumo product and are not treated as independent benchmark inputs.

The current product also includes privacy-aware workflow choices. Web search is optional rather than permanently on, Ghost Mode creates an ephemeral session, stored conversation history can be synchronized under zero-access encryption, and Projects provide persistent context for ongoing work.⁹⁴ Proton Drive integration gives users a route for bringing existing files into the assistant without moving them into a separate third-party AI account.¹⁰

Lumo has a meaningful free entry point. Guest and Free users can access the assistant without a paid subscription; Proton’s current support material says Free includes limited access to the Max model, limited messages and history, limited image generations and one Project.⁴ This is enough to evaluate the interface and privacy model before committing to a subscription.

Lumo Plus raises those limits and adds the fuller persistent-workspace experience: more Max usage, more messages and history, more image generations, unlimited Projects and unlimited Custom Lumos.⁴² The current Italian App Store lists Lumo Plus at €12.99 for one month and €119.99 for twelve months.³ Those are regional in-app purchase prices rather than a universal web tariff, but they provide a current auditable retail reference.

Proton also offers Lumo for organizations through business plans, including Professional and Workspace-oriented offerings, with a trial path on the current business surface.¹¹ Public business pages do not expose a stable numeric price in the currently readable material, so no enterprise price is inferred here.

There is one purchasing detail worth making explicit: a broader Proton subscription does not necessarily equal Lumo Plus. A current user complaint on Proton’s Trustpilot profile concerned the expectation that Proton Unlimited should include unlimited Lumo usage. Proton’s public response clarified that Proton Unlimited includes access to Lumo but not the higher or effectively unlimited usage associated with Lumo Plus.¹² Because that discussion sits on a Proton Mail/company review surface rather than an exact Lumo rating page, it is qualitative plan-clarity evidence only; the Proton Mail aggregate score is not used as a Lumo rating.

The value proposition is therefore different from a generic $20-per-month frontier chatbot. Lumo is cheaper at the current Italian annualized retail reference and offers a useful free tier, but buyers are partly paying for Proton’s privacy and infrastructure choices. Users whose sole objective is maximum benchmark performance per euro should compare output quality on their own workloads rather than assume the privacy premium comes with frontier-leading reasoning.

Privacy is the strongest and best-documented part of Lumo. Proton states that chats are not used to train AI models, that it does not retain server-side prompt logs after processing, that synchronized history is protected with zero-access encryption and that inference runs on Proton-controlled servers in Europe rather than being forwarded to closed third-party AI APIs.⁹¹³⁸

That does not mean the inference itself is end-to-end encrypted or purely on-device. Proton’s technical security model is unusually clear about this distinction: the user’s request travels over an encrypted user-to-Lumo path, but the Lumo GPU server must decrypt the prompt in memory to run inference and generate a response; the server then discards the processing data rather than retaining it as conversation history.¹³⁹ “Zero-access” therefore describes stored synchronized history, not an inability of the inference server to see the prompt while answering it.

This distinction also corrects a misleading description in one current Tom’s Guide privacy article that characterizes Lumo as local AI handled on the user’s device.¹⁴ Proton’s own technical documentation says inference is server-side on Proton-controlled GPU infrastructure. Tom’s Guide remains useful for separate reporting on Proton moving Lumo-related infrastructure toward Germany amid concern about proposed Swiss surveillance-law changes, but its on-device characterization should not be used to describe the architecture.¹⁴

Independent evidence on current capability is thinner than the privacy documentation but no longer negligible. TechCrunch’s June 2026 Lumo 2.0 article verifies the major feature expansion and describes the public version as broadly useful alongside mainstream assistants, without presenting a controlled benchmark.⁶ TechRadar’s current 2.0 coverage similarly supports the direction of travel.⁷ Older PCWorld and The Verge launch coverage remains useful for understanding the original privacy architecture but predates Lumo 2.0; claims from those articles about missing image features or narrower capabilities are historical and should not be projected onto the current product.¹⁵¹⁶

Mobile user evidence is mixed. The current Italian App Store snapshot is 4.1/5 from 39 ratings, while Google Play reports 3.7/5 from roughly 2,840 reviews and more than 100,000 downloads.³⁵ The Android population is much larger and therefore the stronger current user-satisfaction signal; 3.7/5 is respectable but clearly not an unqualified endorsement. The smaller Italian iOS sample is more positive but too small to outweigh the broader Android signal.

WIRED’s August 2026 interview with Proton CEO Andy Yen adds current strategic context around Lumo and Proton’s attempt to make privacy compatible with modern AI products.¹⁷ It is useful for understanding design intent and corporate positioning, not as an independent test of model quality.

Lumo is most compelling for people who want a general-purpose assistant but regard prompts, files and conversation history as sensitive data rather than disposable telemetry. Lawyers, researchers, journalists, executives, security-conscious consumers and teams working with confidential internal material have a clearer reason to choose Lumo than users whose only criterion is leaderboard position.⁹⁸

It also fits users already invested in Proton. Proton Drive integration, the same privacy philosophy and business deployment options can reduce the number of unrelated vendors handling sensitive work data.¹⁰¹¹ The no-account free entry point makes it easy to test without immediately expanding a Proton subscription.⁴

The product is less compelling for users who need a stable, named frontier model with directly comparable benchmark results. Lumo’s routed open-source model stack is part of its architecture, which improves provider independence but makes reproducibility and model-to-model comparisons harder.⁸

Power users should also test difficult reasoning, coding and long-form professional tasks against ChatGPT, Claude, Gemini or other frontier systems before consolidating around Lumo. Current editorial coverage demonstrates that Lumo 2.0 has become substantially more capable; it does not establish that Lumo consistently matches the strongest current assistants across specialist workloads.⁶⁷

Finally, buyers using broader Proton bundles should verify exactly which Lumo entitlement they receive. The distinction between “access to Lumo” and Lumo Plus-level usage is real and has already produced user confusion.¹²