Earlier this month, I wrote about Gemini 3.5 Pro being right around the corner. Two-million-token context window, Deep Think reasoning, multimodal capabilities — it all sounded like a real contender against Claude Opus and GPT-5.6. Turns out the corner was further away than expected.
What Happened
Google has confirmed that Gemini 3.5 Pro won’t reach general availability before July. The model stays in limited Vertex AI enterprise preview for now. This is the second time in a row that Google has missed the delivery window for a model announced at I/O — they had similar delays after I/O 2025.
At I/O in May, Sundar Pichai said: ‘Give us until next month.’ That month is now over.
The Context Makes It Worse
Google says they’re refining the model based on early tester feedback — specifically coding capabilities, token efficiency, and long-task performance. Standard engineering work. But the timing is brutal.
Between June 18 and 24, Google DeepMind lost four senior researchers in six days. Noam Shazeer (Transformer co-author, Gemini co-lead) left for OpenAI. John Jumper (Nobel laureate, AlphaFold lead) left for Anthropic. Jonas Adler (Gemini AI coding lead) and Alexander Pritzel (Gemini pretraining specialist) followed to Anthropic. Alphabet’s stock lost roughly $269 billion in market cap during that stretch.
What It Means
June 2026 is the month where no frontier model launched on schedule and freely available. GPT-5.6 Sol is under government oversight, available only to approved partners. Gemini 3.5 Pro is delayed. Grok 5 was pushed back too.
For developers who’ve been waiting on Gemini 3.5 Pro: Gemini 2.5 Pro with Deep Think (available since June 22) is currently Google’s strongest publicly accessible model. The 3.5 Pro should follow in July — but after two missed deadlines, I’d wait for the actual release day rather than trusting announcements.
Sources: