Google's Gemini 3.5 Pro Misses June Deadline, Slips to July

June 30 came and went without Google shipping Gemini 3.5 Pro to the public. The company confirmed the flagship model is delayed into July, marking its second straight missed launch window — and handing rivals another opening.

Google says the holdup comes down to performance tuning, not a scrapped launch. But the miss lands in a rough month for the company, and it raises the stakes for whatever ships next.

What was supposed to happen

Gemini 3.5 Pro was widely expected to reach general availability by the end of June. Traders were so confident it would slip that a prediction market on Polymarket — asking whether Google's next Gemini Pro model would release by June 30 — closed at 97% on "No," with about $229,697 in total volume.

When a betting market prices your product launch at 3% odds, the news is already out. Google made the delay official rather than rush a flawed release.

Why Google hit pause

The company pointed to a specific problem: excessive token consumption in long, multi-step agentic tasks. In plain terms, the model was burning through too much compute (and running up costs) when handling extended jobs that require many steps of reasoning and tool use.

Google said engineers determined the model needed more validation and optimization on long-horizon performance before a public rollout. Business Insider reported in late June that additional testing work was deemed necessary first.

Why this matters: token consumption is a real cost, both for Google's infrastructure and for customers who pay per token. A model that quietly runs up huge bills on agentic workflows would have drawn immediate criticism — arguably worse than a delay. Shipping a model with documented efficiency problems is the kind of mistake that's hard to walk back.

The features Google is still promising

The expected specs for Gemini 3.5 Pro haven't changed. Google is targeting a 2-million-token context window (roughly the ability to hold and reason over enormous amounts of text at once), a "Deep Think" reasoning mode for tackling harder problems, and frontier-level multimodal capability across text, images, and more.

Deep Think is expected to be gated behind Google's $250-per-month AI Ultra tier, its top consumer plan. For now, the model reportedly remains in a limited enterprise preview on Vertex AI, Google's cloud platform, rather than open to the public.

A bad backdrop for a missed launch

The delay doesn't land in a vacuum. June was, by most accounts, a brutal month for Google's AI standing.

The company lost a string of senior researchers to competitors — several to Anthropic and OpenAI — in a matter of weeks, a talent exodus that reportedly wiped a huge chunk off Alphabet's market value. Google also stood up an internal "coding strike team" to close a perceived gap with Anthropic in AI-assisted software development, and its stock took a sharp single-session hit during the turmoil.

Against that backdrop, a clean, on-time flagship launch would have helped reset the narrative. Instead, Google is heading into July needing Gemini 3.5 Pro to deliver clearly differentiated performance — not just ship.

What to watch next

No firm July date has been confirmed. Google has committed only to a July window, and it's under pressure to make the extra time count.

The bar is higher now because of the delay itself. A model that arrives late and lands as merely competitive won't do much for Google's momentum. One that arrives late but posts standout results on reasoning, coding, and efficiency could flip the story quickly — especially given how much of June's coverage focused on Google's stumbles rather than its models.

There's also a competitive clock ticking. OpenAI and Anthropic have both been active, and every week Gemini 3.5 Pro stays in preview is a week rivals can use to lock in developers and enterprise customers.

The bottom line

Google's delay is defensible on the merits — shipping a model that guzzles tokens on agentic tasks would have been a worse look than waiting. But defensible isn't the same as harmless. This is the second consecutive missed window, and it comes during the company's hardest stretch of the year.

If you're planning to build on Gemini 3.5 Pro, don't lock in timelines yet. Wait for the actual July release and check the token-efficiency numbers on long agentic runs before you commit — that's the exact metric Google says it went back to fix.