Google slips Gemini 3.5 Pro to July as Shazeer and Jumper exit DeepMind

Gemini 3.5 Pro, Google's flagship frontier model, has slipped from its June I/O window into July 2026, according to Business Insider, days after Noam Shazeer decamped to OpenAI and John Jumper to Anthropic. The 48 hours separating those two exits also took roughly $225 billion off Alphabet's market cap, with shares falling as much as 7.2% intraday Monday before closing down about 5%, the worst session for the stock since February.

The schedule itself is the tell. Gemini 3.1 Pro shipped widely in February; a July 3.5 Pro puts roughly five months between Pro releases, a cadence in which Anthropic has now pushed two Claude Opus updates and previewed Mythos, a new long-horizon model class. Business Insider reports the slip reflects ongoing tuning against early user feedback from Antigravity and LMArena. Sundar Pichai owns the calendar slip; the talent slip is the part Wall Street can't price.

Shazeer is the harder loss to absorb. He co-authored 2017's "Attention Is All You Need," the paper from which essentially every modern LLM descends, left Google in 2021 (alongside Daniel De Freitas) over the company's reluctance to ship LaMDA, founded Character.AI, and was reabsorbed in August 2024 as VP of engineering on Gemini. That round-trip was supposed to be the proof point: that DeepMind could retain a founder-class researcher inside a $2-trillion bureaucracy. It held for under two years.

Jumper's exit lands differently. His AlphaFold work shared the 2024 Nobel Prize in Chemistry, and his move to Anthropic, run by Dario Amodei, signals that scientific prestige now routes through the smaller labs. That's a structural reversal of the talent gravity that defined Google Research for a decade.

The shipped half of the story is genuinely strong. Gemini 3.5 Flash, already in the wild, posts 76.2% on Terminal-Bench 2.1, 1656 Elo on GDPval-AA, 83.6% on MCP Atlas, and 84.2% on CharXiv Reasoning, with output-token throughput Google claims is roughly 4x other frontier models. Flash is competitive. The problem is the Pro-tier story, and the gap is becoming legible to the leaderboards: Fortune notes that Gemini 3.5 Flash and 3.1 Pro both sit outside the top five on multiple public boards, behind not only OpenAI and Anthropic but Chinese entrants Zhipu AI and MiniMax.

That last detail is what the 5% drawdown is really pricing. The 2021 LaMDA episode hardened into Google lore as the moment the company learned to ship; the 2024 reabsorption of Shazeer was meant to close the loop. Bloomberg, CNBC, Fortune, and the Wall Street Journal are now writing variations on the same question, which is whether DeepMind can hold its frontier position without the researchers who defined it. A July Pro release will arrive into a market that has already started answering.

Sources