Google shipped Gemini 3.8 Live and Live Extended Thinking on Tuesday, a day after Apple released Siri AI. Both aim at home and personal users, both are free to start. And hindsight being 20/20, it appears Gemini and Siri AI (i.e. built with Google’s Gemini) are losing to Claude and Codex on performance. I would argue Gemini is chasing a different axis entirely. A model that is faster, auditory, and personal. And as it stands for the time being, there is no automatic delegation between them.

Google’s announcement leads with the top spot in Artificial Analysis’ Speech to Speech Quality Index, then a model that switches between 97 languages mid-conversation while narrating its progress aloud. Apple’s opens on expressive voices, self-punctuating dictation, and replies that reflect how you usually write to each recipient. Neither claimed its model reasons better. The features on display are ones Gemini could already do, given access to your private data. What is new is the voice, and what is new is the pipeline to your data.

The axis they left has a scoreboard too, and Google did not lead with it. Artificial Analysis also runs AA-Omniscience, whose hallucination rate — lower is better — measures whether a model guesses or admits it does not know. Gemini 3.8 Live is not scored on it; its sibling 3.8 Flash guesses on 55% of what it cannot answer correctly, against 39% for Claude Sonnet 5. I suspect that is not incidental. Google’s own illustration of fluency assumes an active human on the other end, and when it comes to coding I don’t see Gemini as the tool professionals reach for. What it specializes in is ‘this is the draft for George in accounting’, for you to glance over before hitting the send button. That glance is a stopgap, and it is the price you pay for free access.

Neither is charging you an entrance fee, and both Gemini and Siri AI kept that gate open. Google gives Extended Thinking to everyone in Gemini Live while its Workspace equivalents sit behind Google AI tiers. Apple already meters Apple Intelligence in the Home app through iCloud+, and says expanded Siri AI access “will be available for a fee in the future” — for now it runs free under daily limits. Claude and ChatGPT sit at the other end. OpenAI files its free plan under “best for trying out ChatGPT”, and what both companies advertise sits on the tiers above. There is a saying that the first $20 month is the real trial, not the free conversation.

Gemini and Siri AI give away more than Claude and ChatGPT do, and last I checked, paying Google or Apple more did not return as much as paying the other two. Apple is not competing on the reasoning axis. It describes its new foundation models as “custom-built in collaboration with Google and its Gemini models”. What Apple is assembling is an App Store for AI, and Apple Intelligence is simply the first title stocked. The ChatGPT extension has been available for some time, and Claude has been found working in shipped iOS 27 frameworks, though Apple has not opened the entitlement and nothing is announced. They take care of the free half. The other half you buy yourself, from whoever the store decides to stock.