Google's 7 Hidden Gemini Models Exposed; Intel Closes In on Apple Chip Deal

Google is testing seven specialized Gemini Live voice models hidden from users as Intel closes in on an Apple chip deal that would validate its foundry transformation.

🤖 Hidden Gemini Live Selector Reveals 7 Unreleased Voice AIs Ahead of Google I/O

Decoded: A hidden model selector in Google App v17.18.22, uncovered by Forbes on May 12, reveals that Google is actively testing seven previously unreported AI models for Gemini Live voice conversations ahead of Google I/O 2026, which begins May 19. The models — including ones codenamed "Capybara" and "Nitrogen" — exhibit measurably different capabilities in testing. Four models can access real-time user location and provide live weather data; three cannot. One model, "Capybara," identifies itself as "Gemini 3.1 Pro" rather than the standard Flash Live model when queried. Several models demonstrate different memory behaviors, with some recalling conversation details more reliably than others. The switchable model infrastructure is already operational but hidden behind a server-side flag, indicating active road-testing before public launch. Google has not officially announced the models or confirmed the selector. (Forbes, May 12, 2026)

Why it matters: The leak signals Google is preparing to offer Gemini Live users a meaningful choice between specialized voice AI configurations — optimized for factual research, personalized memory, or real-time context. The fact that at least one model already identifies as Gemini 3.1 Pro, Google's flagship reasoning model, suggests the company is testing whether its highest-capability tier can operate in real-time voice mode at scale before announcing it at I/O. For Alphabet (GOOGL), a differentiated voice model lineup would directly compete with OpenAI's GPT-Realtime-2 voice API, announced May 11 — both companies are racing to own the primary interface through which users access AI assistants.

🗄️ Intel Nears Apple Chip Deal as AI Revenue Surge Drives 500% Stock Run

Decoded: Intel's stock has risen nearly 500% over the past 12 months and approximately doubled in the past month, driven by stronger-than-expected quarterly results, booming AI demand, and reports that Intel's foundry business is approaching a deal to supply custom chips to Apple — which currently manufactures all of its A-series and M-series chips at TSMC. CEO Lip-Bu Tan confirmed in the company's April earnings call that AI demand was driving measurable revenue growth and that Intel's 18A process node was on track for commercial readiness. An Apple foundry agreement would be the first time Apple has manufactured custom silicon outside TSMC since returning chip production to the island in the early 2010s. Intel and Apple have not confirmed the reports. (Reuters, CNBC, May 11, 2026)

Why it matters: A confirmed Apple foundry deal would be the most consequential commercial validation of Intel's transformation into a foundry-as-a-service competitor. Apple ships approximately 250 million iPhones per year; a meaningful production share moving to Intel's 18A fabs — even as a secondary source — would immediately de-risk Intel's foundry revenue and justify the multi-billion dollar domestic fab investments the company has made. For TSMC (TSM), a production split would reduce its Apple concentration risk, which accounts for an estimated 25% of TSMC revenue. The 500% stock run reflects the market pricing in a scenario where Intel successfully transitions from struggling chipmaker to credible U.S. foundry at a moment when semiconductor sovereignty is a stated national priority.

Stay decoded. See you tomorrow.

— The Get AI Decoded Team