Google said Gemini 3.5 Pro was 'next month'. The month's over.

At I/O, Google shipped Gemini 3.5 Flash — and it's genuinely good for its tier, default across the Gemini app and Search. Then it announced 3.5 Pro, didn't release it, and said it was in internal use with a rollout 'next month'. The split is the story. The model you can touch is the cheap one. The model with the big numbers attached is the one you can't. By late June there's no consumer access, no AI Studio listing, no stable API, no model card. Pro exists the way a reservation exists: someone wrote it down, and you still can't sit at the table.

Here's the part the breathless coverage skips. The numbers everyone is quoting for 3.5 Pro — a 2-million-token context, a 'Deep Think' mode — are not Google's confirmed specs. They're trade-press expectations, repeated until they sound official. That's how an unreleased model accumulates a fake résumé: a launch-adjacent blog floats a figure, an aggregator restates it without the hedge, and three weeks later 'Gemini 3.5 Pro has a 2M context' is treated as settled. It isn't. Google's Flash materials list about a 1M context. Pro's real numbers arrive with a model card, and there isn't one — until Google publishes it, every spec attached to Pro is somebody's educated guess, not a spec.

Be fair about what kind of criticism this is. Google ships real models — Flash is proof, released on the day and working. Holding the heavier Pro model back to harden it while getting the fast one into millions of hands is a reasonable sequencing decision, arguably the right one. The complaint isn't 'Google can't ship'. It's narrower and it's about communication: don't put a 'next month' on a flagship from a keynote stage unless you're prepared for 'next month' to be quoted back at you when it passes. The audience hears a date. The lawyers wrote a vibe. The gap between the two is where trust leaks.

There's also a competitive clock running that makes the slip more than a calendar footnote. The frontier is a weekly news cycle right now — OpenAI and Anthropic are shipping and re-shipping, and every week Pro stays in the lab is a week the 'best Gemini you can actually use' is the mid-tier Flash, not the flagship. That's a fine position to hold for a sprint and an awkward one to hold for a quarter. None of that is visible from the spec rumours; it's the part that actually matters for anyone choosing a model to build a product on this summer. The question isn't whether Pro is impressive on paper — it's whether you can call it from an API before your roadmap needs it, and right now you can't.

If you're making a decision: build on Flash, which exists and is strong for its class, and treat Pro as a maybe. Don't wire a roadmap to a 2M-token context that no one at Google has confirmed for a model no one outside Google has used. When the model card drops — real benchmarks, a stable endpoint, an actual availability date that's a date — that's when Pro becomes a thing you can evaluate instead of a thing you can quote. Until then, the most honest sentence about Gemini 3.5 Pro is short: announced, not shipped, and a month past its own 'next month'.