{"slug": "introducing-lumo-2-0", "title": "Introducing Lumo 2.0", "summary": "Proton launched Lumo 2.0, a major update to its zero-access encrypted AI assistant, introducing advanced reasoning models, multimodal capabilities including image recognition and generation, enhanced web search, and user-controlled memory. The update, built on new architecture and protected by Swiss privacy laws, aims to bring frontier AI capabilities to over 10 million users while maintaining privacy guarantees.", "body_md": "Last year we launched Lumo, a zero-access encrypted AI assistant that never logs your conversations or trains on your data. Since then, more than 10 million people have started using it, and we’ve shipped major updates, including [custom conversation styles](/blog/lumo-1-2), dedicated [encrypted project spaces](/blog/lumo-1-3), more powerful models, and more.\n\nToday we’re announcing Lumo 2.0, the most significant change since launch. We’ve rebuilt Lumo on a new architecture with capabilities that bring it to the frontier of what AI can do. As always, it runs on Proton’s fully European infrastructure, protected by [Swiss privacy laws](/blog/switzerland) and [zero-access encryption](/blog/lumo-security-model).\n\n## Advanced reasoning models\n\nLumo 2.0 is our biggest leap in capability yet. On the [Artificial Analysis Intelligence Index(new window)](https://artificialanalysis.ai/), an independent benchmark combining 10 evaluations across agents, coding, scientific reasoning, and general knowledge, Lumo 2.0 Lite scores 127% higher than Lumo 1.4, and Lumo 2.0 Max scores 240% higher.\n\nLumo 2.0 also introduces fast and reasoning [modes](/support/lumo-models) to make everyday interactions feel faster while giving more demanding tasks additional depth. **Fast **prioritizes speed and\n\n**Thinking** is optimized for more complex, multi-step reasoning — a capability that didn’t exist in Lumo 1.4.\n\nLumo 2.0 responds to everyday queries **up to 76% faster than Lumo 1.4**. More complex tasks requiring deeper thinking now benefit from a visible thinking state, letting you see how Lumo is working through the problem in real time.\n\n## Image recognition and generation\n\nLumo 2.0 is now multimodal, able to process both text and [images](/support/lumo-create-edit-image). Upload an image to analyze, create visuals from a prompt or a rough sketch, or edit existing images, all in the same conversation.\n\nLumo’s zero-access encryption applies to image processing too. Images you upload and the images Lumo generates are stored so that no one, not even Proton, can access them.\n\n**Image generation**\n\nDescribe what you want and Lumo creates it. Type a prompt, refine it in the same conversation, and iterate until it looks exactly right. From a logo concept to a travel poster, start with words and end with an image.\n\n**Image editing**\n\nUpload an image and tell Lumo what to change. Swap backgrounds, adjust colors, remove unwanted objects, or completely re-imagine a scene.\n\n**Image analysis**\n\nUpload a chart, a document, a screenshot, or a photo and ask Lumo anything about it. Lumo identifies what’s in the image, reads the data, and responds with detail.\n\n**Sketch to image**\n\nUpload a rough sketch and describe what you want it to become. Tell Lumo the style, the mood, the details, and watch it take shape.\n\n## Powerful new web search capabilities\n\nCompared to Lumo 1.4, Lumo 2.0 has far stronger web search. When a question calls for current information, Lumo pulls live results from across the web and cites its sources, so you can verify answers and dig deeper. It stays current on recent news and events, draws on live financial data, and can even pull in weather forecasts. The result is more accurate, more complete answers with fewer hallucinations.\n\n## Deeper context with memories\n\nLumo 2.0 combines user-controlled memory and Projects to move beyond one-off questions toward collaboration that builds over time.\n\n[Memory](/support/lumo-memory) lets Lumo learn your preferences, working style, and ongoing context, so every conversation starts smarter than the last. You decide what it retains, what it forgets, and what it never learns in the first place.\n\nThe context window is now twice as large, so Lumo holds longer conversations coherently and reasons across longer documents and bigger datasets. For work that spans multiple sessions, Projects go further: Each is a dedicated encrypted workspace that keeps your chats, files, and instructions together.\n\nEvery piece of context is protected by zero-access encryption, the same architecture that secures over 100 million Proton Mail and Proton Drive users.\n\n## Custom Lumos\n\nLumo 2.0 also introduces [Custom Lumos](/support/custom-lumos), purpose-built assistants tailored to specific tasks. Build a writing assistant that always drafts in your tone, or a research assistant that structures answers the way you need them. You define the instructions once, and the Custom Lumo follows them every time, no re-explaining required.\n\n## Private AI for businesses\n\nMost AI tools create a new category of risk for businesses. Employee queries become training data. Confidential documents can be compromised. And your data sits on infrastructure subject to US law — with recent events already showing how fast access can vanish for international users.\n\n[Lumo for Business](/business/lumo) is built for organizations that can’t afford those risks. Every conversation is zero-access encrypted, never logged, and never used to train future models. Administrative tools let you manage your team’s access, and your data stays on independent European infrastructure, meaning access to Lumo cannot be subject to US Executive Orders and user data is not subject to American data collection requests.\n\n## Intelligence without surveillance\n\nWith Lumo 2.0, you no longer have to choose between advanced [AI and privacy](/lumo/ai).\n\nPeople now share more with AI systems than they ever shared with any other technology, and the dominant providers are normalizing a future where your personal data is the price of participation. Conversations are getting [mined for training](/blog/family-photos-ai-risks). Behavior is being used as a [signal for profiling](/blog/ai-privacy-concerns). [Ads are arriving](/blog/chatgpt-ads) inside the most popular assistants. And control over all of it is now concentrated in the hands of [a few US tech companies](/blog/private-ai) (many with government and defense contracts) with increasing commercial pressure to monetize your attention.\n\nWe believe AI should empower people, not extract from them. As it becomes infrastructure for work, creativity, and communication, privacy can no longer be optional. Lumo brings Proton’s privacy standards to AI: zero-access encryption, no logs, no data sharing, and no training on your conversations. And because Lumo is fully open source, anyone can inspect the code, verify the encryption, and confirm it works exactly as we say.\n\n## Lumo 2.0 is live now\n\n**Free:** for private AI with core capabilities.__Sign up__(new window)**Lumo Plus:** Unlimited chats, Projects, advanced image generation, and priority access to the fastest models.**Lumo Professional:** Premium features for teams, enabling secure collaboration without data leak risks.\n\nWe are shaping a future where AI works for you. Share your thoughts on [X(new window)](https://x.com/asklumo) and [Reddit(new window)](https://www.reddit.com/r/lumo/).", "url": "https://wpnews.pro/news/introducing-lumo-2-0", "canonical_source": "https://proton.me/blog/lumo-2", "published_at": "2026-06-30 10:32:56.101073+00:00", "updated_at": "2026-06-30 10:32:58.855828+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-products", "ai-safety", "ai-ethics", "ai-infrastructure"], "entities": ["Proton", "Lumo", "Artificial Analysis Intelligence Index", "Switzerland"], "alternates": {"html": "https://wpnews.pro/news/introducing-lumo-2-0", "markdown": "https://wpnews.pro/news/introducing-lumo-2-0.md", "text": "https://wpnews.pro/news/introducing-lumo-2-0.txt", "jsonld": "https://wpnews.pro/news/introducing-lumo-2-0.jsonld"}}