Inside Apple’s labs, engineers are testing multiple large language models for Siri that match, and in some cases surpass, ChatGPT’s latest benchmarks. Internal builds run from 3 billion to a hefty 150 billion parameters, dwarfing the 3-billion-parameter model that powers the first wave of Apple Intelligence features.
Bigger brains, cloud power
To reach that size, Apple shifted from on-device processing to cloud servers. The move frees Siri from iPhone silicon limits and lets the assistant handle longer prompts, nuanced follow-ups, and multi-step tasks. However, larger models mean higher energy costs and tougher privacy safeguards, both of which executives must balance before a public debut.
Hallucinations remain a hurdle
Despite impressive accuracy, the most advanced model still “hallucinates” more often than Apple’s quality bar allows. False facts pose brand-risk for a company that sells trust alongside hardware. Engineers are refining guardrails, but insiders say the error rate continues to spark debate at the highest levels.
Leadership split slows the timeline
Sources describe a philosophical rift within Apple’s senior ranks. Some leaders push to ship the cloud-based Siri quickly and iterate in public. Others, wary of reputational damage, argue for keeping the chatbot under wraps until hallucinations drop further. This internal tug-of-war has already trimmed the list of Apple Intelligence announcements expected at June’s WWDC.
What users might see first
Even with delays, hints of the upgrade should surface. Apple is likely to unveil modest Siri boosts—such as better context retention or hands-free shortcuts—while previewing a broader redesign of iOS 19’s interface. Incremental releases would let Apple prove reliability at each step, easing skeptics and regulators alike.
Why it matters
Apple once lagged rivals in generative AI, yet these prototypes signal rapid progress. If the company tames hallucinations and resolves leadership concerns, a ChatGPT-class Siri could land on hundreds of millions of devices—reshaping how users search, message, and create on Apple platforms. For now, the world will watch WWDC for any glimpse of the smarter assistant waiting in the wings.
