1
What model is it closest to in intelligence? Pro, flash, Gemma 31?
Apple's custom Siri model is firmly in frontier Pro territory — far above Flash and roughly 40x larger than Gemma's biggest model.
Why it matters: The gap matters because Flash and Gemma trade raw intelligence for speed or openness; Apple paid ~$1B/year specifically to avoid that compromise.
- The custom model has 1.2 trillion total parameters using a mixture-of-experts (MoE) architecture — only a relevant subset fires per query, but total capacity dwarfs any Flash or Gemma variant.
- Gemma 4's largest open model tops out at ~27–31B parameters; Gemini Flash is optimized for speed and cost, not maximum reasoning — the 1.2T custom model is in a different class from either.
- Gemini 3.1 Pro currently leads Google's public lineup on hard benchmarks (94.3% GPQA Diamond, 80.6% SWE-bench Verified); the Apple model's scale suggests it sits at Pro level or above — possibly matching an unreleased Ultra-class variant.
- Google has not disclosed the active parameter count per query for the custom model, so a precise 'Pro vs. Ultra' ranking is genuinely uncertain — but Flash and Gemma are ruled out by the numbers alone.
Room for disagreement
- Google hasn't released benchmark data for the custom Apple model specifically, so some analysts argue it could be a stripped-down frontier model tuned for assistant tasks rather than a true Pro-class reasoner — meaning real-world intelligence could land closer to a high-end Flash than a full Pro.
AI-generated with web search — may be wrong. Check the linked sources.
