Apple’s Siri received its most comprehensive overhaul in 2025-2026 with the introduction of what Apple calls “Siri 2.0.” The changes go far beyond a new interface — this is a fundamentally different assistant powered by advanced language models and deep system integration. Here is what changed and what it means for daily use.
The Core Intelligence Upgrade
Siri 2.0 is built on Apple’s on-device foundation models combined with cloud processing for complex queries. The assistant now understands context across your device — it knows what you are working on in Mail, what calendar conflicts you have, and can take actions within applications, not just answer questions.
The language understanding improvement is dramatic. Previous Siri struggled with natural conversation flow — you had to phrase commands precisely. Siri 2.0 handles conversation, follow-up questions, and complex multi-step requests that would have confused the old version entirely.
App Actions and System-Wide Integration
Siri can now take actions within third-party applications with user permission. You can ask Siri to “send a message to David on WhatsApp” or “add this PDF to Things” or “find the receipt from my last flight in Outlook.” The App Intents framework that enables this requires developer support, but major apps are adding Siri integration rapidly.
Writing Assistance
Siri 2.0 includes the full Writing Tools suite: proofreading, rewriting in different tones, summarization, and text generation. This works in any text field system-wide through the share sheet or keyboard. The quality matches ChatGPT and Claude for most common writing tasks.
On-Device vs Cloud Processing
Simple requests (setting timers, controlling HomeKit, basic queries) process entirely on-device on M-series chips and A17 Pro or later. Complex requests that require more processing are sent to Apple’s Private Cloud Compute infrastructure. Apple maintains that no user data is stored or accessible to the company.
What Still Does Not Work Well
Siri 2.0 still struggles with highly specialized knowledge — it is not a research tool. Complex scheduling across multiple calendars with conflicting constraints still requires human judgment. The assistant is significantly better but not equivalent to using a language model directly for complex research or analysis tasks.