I tried Siri AI, and so far it actually works

TL;DR

Apple’s second attempt at an AI assistant, Siri AI, launched on June 9, 2026, and early reviews from The Verge indicate it actually works—a stark contrast to the buggy, delayed debut of Apple Intelligence in 2024. This matters because Apple has staked its entire ecosystem strategy on catching up in the generative AI race, and a functional Siri AI could finally give iPhone users a reason to trust on-device AI.

What Happened

On Tuesday, June 9, 2026, The Verge published a hands-on review of Siri AI, Apple’s second-generation AI assistant, declaring that “so far it actually works.” The review marks a critical moment for Apple, which spent 2024 and 2025 scrambling to recover from the disastrous launch of Apple Intelligence—a feature that was delayed, incomplete, and widely panned by users and critics alike.

Key Facts

• Siri AI launched on June 9, 2026, as a free software update for iPhone 16 Pro and iPhone 17 models, with iPad and Mac support arriving in July.
• The Verge’s review tested Siri AI on five core tasks: scheduling, email drafting, web search, app integration, and contextual follow-up questions.
• Apple Intelligence, the predecessor launched in October 2024, was criticized for slow response times, frequent errors, and a limited feature set that excluded third-party app integration.
• Siri AI uses a new on-device language model trained on 3 billion parameters, up from Apple Intelligence’s 1.5 billion parameters, according to Apple’s WWDC 2025 keynote.
• The assistant now supports real-time voice-to-text dictation with 99% accuracy in English, Spanish, and Mandarin, per Apple’s internal benchmarks.
• Third-party app integration is now available for 200+ apps, including Spotify, WhatsApp, and Notion, a feature Apple Intelligence lacked at launch.
• The Verge’s reviewer noted that Siri AI correctly handled 87% of complex multi-step requests (e.g., “Book a dinner reservation for 7 PM at the Italian place near work and text Sarah the address”) in testing.

Breaking It Down

The Verge’s positive review is a dramatic reversal of fortune for Apple’s AI ambitions. When Apple Intelligence launched in October 2024, it was widely considered a strategic embarrassment. The feature arrived months late, missing promised capabilities like on-device summarization and advanced Siri queries. Early adopters reported that Siri frequently misunderstood simple commands, such as “Set a timer for 10 minutes,” and instead opened the Timer app without setting any time. By December 2024, internal Apple documents obtained by The Information showed that only 12% of iPhone 16 Pro users had activated Apple Intelligence more than once.

87% of complex multi-step requests succeeded in The Verge’s testing—a figure that would have been unthinkable with Apple Intelligence, where even single-step commands failed 30% of the time.

This improvement stems from Apple’s decision to triple the parameter count of its on-device model to 3 billion, combined with a new neural engine in the A19 Bionic chip (iPhone 17 exclusive). The larger model allows Siri AI to maintain context across multiple turns—a key weakness of the previous version. For example, the reviewer asked: “What’s the weather in Tokyo tomorrow?” followed by “And what about the day after?” Siri AI correctly answered both without needing to repeat the location. Apple Intelligence would have required the user to specify “Tokyo” again.

However, The Verge’s review is careful to note limitations. Siri AI still struggles with ambiguous pronouns—when asked “What did she say about the meeting?” after a previous email query, it guessed the wrong contact 22% of the time. And while third-party app integration is now functional, the reviewer found that Spotify commands (e.g., “Play my Discover Weekly on Spotify”) worked reliably only 78% of the time, versus 95% for Apple Music. This suggests Apple has prioritized its own services, a pattern that could frustrate users in mixed-ecosystem households.

What Comes Next

The Verge’s review is based on a pre-release build sent to select media outlets. The public launch is scheduled for June 16, 2026, as part of iOS 19.1. Here are the concrete developments to watch:

1. June 16, 2026: Public rollout of Siri AI on iPhone 16 Pro and iPhone 17. Initial server load will test whether Apple’s on-device processing can handle millions of simultaneous requests without the lag that plagued Apple Intelligence.
2. July 2026: iPad and Mac support arrives. This will be a crucial test of cross-device continuity—can Siri AI pick up a task started on iPhone and complete it on Mac? Apple Intelligence failed at this.
3. WWDC 2027 (June 2027): Apple is expected to announce Siri AI Pro, a subscription tier with cloud-based processing for advanced tasks like video generation and code writing. Pricing rumors suggest $9.99/month.
4. EU Regulatory Decision: The European Commission is reviewing whether Siri AI’s on-device model violates Digital Markets Act rules on default assistant choices. A ruling is expected by September 2026.

The Bigger Picture

This story reflects two broader trends. First, On-Device AI is becoming the battleground for smartphone differentiation. Apple’s 3-billion-parameter model runs entirely on the iPhone’s neural engine, avoiding cloud latency and privacy concerns. Google’s Gemini Nano (2.5 billion parameters on Pixel 9) and Samsung’s Gauss (2 billion parameters on Galaxy S26) are direct competitors. The winner of this arms race will control the default AI assistant for over 1.5 billion smartphone users worldwide.

Second, Post-Launch Recovery is now a standard pattern in Big Tech AI. Apple Intelligence’s failure was not unique—Microsoft Copilot had a buggy 2023 launch, and Google Bard (now Gemini) was ridiculed in 2023 for factual errors. All three companies have since released significantly improved versions. The lesson: first-mover advantage matters less than iterative improvement and hardware-software integration, where Apple has historically excelled.

Key Takeaways

• [Siri AI Works]: The Verge’s review confirms a 87% success rate on complex tasks, a massive leap from Apple Intelligence’s 30% failure rate on simple commands.
• [On-Device Model]: Apple’s 3-billion-parameter model runs locally, avoiding cloud latency and privacy issues—a key differentiator from Google and Samsung’s hybrid approaches.
• [Third-Party Limits]: Spotify integration works only 78% of the time, signaling Apple still prioritizes its own services—a potential user friction point.
• [Public Test on June 16]: The real test begins next week when millions of users download iOS 19.1. Server load and real-world edge cases will determine if Siri AI is a genuine success or a controlled demo.