Sundar Pichai went on the I/O 2026 keynote stage on May 19 and said, of Gemini 3.5 Pro, “give us until next month to get it to you.” Next month is now this month. We are at June 16. The Pro tier is still in internal Google use and a limited Vertex preview accessible to a handful of enterprise design partners. The 3.5 Flash tier shipped the same day as the keynote and has spent the last four weeks doing essentially everything Google needs Gemini to do in public, including the agentic Astra demos at Google I/O Connect last week.

The Pro pitch on stage was a 2-million-token context window, a Deep Think reasoning mode that runs longer chains for harder problems, and the absorption of the use cases Google had previously routed to the Ultra tier. The benchmark slides at I/O suggested parity with or modest leads over Claude Opus 4.7 and GPT-5.5 on long-context retrieval and on the harder reasoning evals. None of that has been independently verified at scale, because the access has been narrow enough that no third party has had a representative sample to test against.

The most likely reason for the slip is the one Google has not stated, which is the Deep Think inference cost. Long-chain reasoning at 2-million-token context is expensive per call. The pricing question is the gating question for a public release, not the model-quality question, because Google does not want to ship a tier that loses money on every call and then either eat the margin or reprice within six weeks. The internal debate, if you believe the people in the Mountain View building who talk to reporters off the record, is whether to ship at parity with Opus 4.7 pricing and accept thin margins, or to add a Pro Plus tier above Pro that prices Deep Think separately and lets Pro itself remain in a sane unit-economics zone.

The competitive pressure is on the calendar. Anthropic’s Mythos 5 is offline because of a Commerce directive, which removes the most-feared comparison point for two to six weeks. OpenAI’s GPT-5.5 has been GA on Bedrock since the start of the month and is the active enterprise default until Google ships. Every week Pro slips is a week of Vertex enterprise pipeline that gets pulled into renewal conversations from a weaker position. Pichai’s “give us until next month” worked once. It does not work twice.

googlegeminigemini-3-5-progemini-3-5-flashdeep-thinkvertex-aisundar-pichaimodel-release-cycles