Skip to content
News Model release

Gemini 3.5 Pro slips to July as Google misses Sundar Pichai's I/O commitment

· by Pondero Newsdesk

The short version

Google has pushed Gemini 3.5 Pro's general availability to July 2026, breaking a public June window Pichai gave developers at Google I/O and arriving as two of Google's most decorated AI scientists leave for OpenAI and Anthropic.

Gemini 3.5 Pro slips to July as Google misses Sundar Pichai's I/O commitment

Google has pushed Gemini 3.5 Pro's general availability to July 2026. The slip breaks a public timeline Sundar Pichai set at Google I/O on May 19, where he told developers to "give us until next month" for the full release. That month has passed and no GA entry appears in the Gemini API changelog as of June 25. The delay lands in the same week that Google lost two of its most prominent AI researchers to rival labs.

What happened

Business Insider reported on June 24 that Gemini 3.5 Pro slipped to July, citing a person familiar with the matter. Crypto Briefing and Startup Fortune independently confirmed the timeline on the same day. Google is gathering feedback from early testers through its Antigravity platform and LMArena, and is incorporating lessons from the Gemini 3.5 Flash rollout, including reported concerns about token use in long-horizon agentic tasks.

The model has been in limited Vertex AI enterprise preview since I/O, featuring a 2-million-token context window and a Deep Think reasoning mode gated to Gemini Ultra subscribers at $250 per month. Pricing has been reported at $15 per million input tokens, though Google has not confirmed that figure publicly.

Gemini 3.5 Flash, the lighter model in the same family, went generally available on May 19, per the official Gemini API changelog. Gemini 3.5 Pro has no corresponding entry. A Google spokesperson declined to comment on the timeline, per Business Insider.

The talent context that makes this harder to dismiss

A routine schedule adjustment lands differently when set alongside what else happened this week. On June 18, Noam Shazeer announced he is leaving Google for OpenAI. Shazeer is a co-author of "Attention Is All You Need," the 2017 Transformer paper, and had been a co-lead of the Gemini model program. Google had paid approximately $2.7 billion in August 2024 to bring Shazeer back through the Character.AI licensing deal. That arrangement lasted under two years.

Two days later, John Jumper announced his departure from Google DeepMind to join Anthropic. Jumper led the AlphaFold 2 project and shared the 2024 Nobel Prize in Chemistry with DeepMind CEO Demis Hassabis. His work produced a database of protein structure predictions now used by more than 2 million researchers across 190 countries. Neither researcher has a named successor at Google.

Startup Fortune reported that Alphabet shares fell 5% on June 22, wiping approximately $225 billion in market value, as investors absorbed the back-to-back departures alongside the model delay news.

The two situations are not directly linked. Shazeer and Jumper did not build Gemini 3.5 Pro in the way that a single engineer owns a feature. A large model program has dozens of contributors, and Google DeepMind retains substantial research capacity. The delay is attributed to quality work on agentic task performance. But context shapes how enterprise buyers interpret a calendar slip, and that context is currently unfavorable for Google.

Why this matters for teams choosing a model stack

The practical pressure falls on organizations now making second-half 2026 infrastructure decisions. Procurement calendars and security reviews take weeks. Developers standardize habits around whichever tool is available when a budget clears, and a model that arrives late has to overcome defaults that formed without it.

Gemini 3.5 Pro was positioned as Google's answer to Claude Opus 4.8 and GPT-5.5 for complex reasoning and long-context coding work. Enterprise buyers who had been waiting to evaluate it before committing to a platform now face another wait. Claude and GPT-5.5 do not face the same timeline pressure right now. For teams running competitive evaluations before Q3 spend locks in, July is not abstract: it is the deadline.

Operators who had provisioned Vertex AI access for early Gemini 3.5 Pro testing should watch for the GA announcement on the Gemini API changelog. Nothing in the current delay language suggests the model is in trouble technically. Google's stated reason is iterative feedback integration, which is the work you want done on a model built for long-horizon agent tasks. The question is whether "July" means early or late July, and whether the feature set at GA matches what Pichai described at I/O.

What to watch next

The Google DeepMind blog and the Gemini API changelog are the two concrete signals to track. A GA announcement before July 15 with the Deep Think reasoning mode intact would contain the delay narrative. A slip past mid-July, or a GA release that omits promised capabilities, shifts the story from a scheduling adjustment to a product problem. Google naming a successor structure to Shazeer's Gemini co-lead role would be a separate signal worth watching for anyone assessing research continuity inside Google's AI division.

Sources