Model Releases · JUNE 24, 2026
Gemini 3.5 Pro Is Six Days From Its June Window With No Ship Date
Google's flagship sits in limited Vertex AI preview as the June GA window narrows. Prediction markets give it 50–55% odds — and the field has moved.
As of June 24, with six days left in the window Sundar Pichai announced from the Google I/O stage, Gemini 3.5 Pro remains a limited Vertex AI preview with no public release date, no benchmark card, and no pricing. The DeepMind models page still lists it as "coming soon." Prediction markets, per TechJournal's June 23 status check, give it 50–55% odds of landing by June 30.
That's a thin margin for a model Google framed as its flagship answer to a competitive cycle that has, in the intervening five weeks, lapped it twice.
Rewind to May 19. Pichai used I/O to ship Gemini 3.5 Flash, which TechCrunch's Rebecca Bellan described as Google's pivot from chatbot to agent infrastructure and which Google itself called its "most powerful coding and agentic AI model yet." Pro was held back. Pichai asked the audience to "wait another month." That month is almost up.
What the public knows about 3.5 Pro comes from Google and a June 6 TechTimes writeup: a 2-million-token context window, a reasoning mode called Deep Think, frontier multimodal understanding. None of it's independently verified. Epoch AI's MRCR v2 retrieval benchmark, the closest thing to a neutral yardstick, scores Gemini 3.1 Pro at 84.9% and 3.5 Flash at 77.3% at 128k tokens. There's no 3.5 Pro row. The Thomson Reuters testimonial Google features on the DeepMind page, in which CTO Joel Hron praises "a clear step forward in handling structured business tasks that require precision and consistency," refers to the 3 Pro generation, not 3.5 Pro.
Meanwhile the field moved. On June 15, Anthropic's Claude Fable 5 posted an ECI score of 161, one point ahead of GPT-5.5 Pro and Anthropic's first ECI lead in over a year. Three days earlier it had scored 87% on FrontierMath Tiers 1–3 and 88% on Tier 4. On June 22, Epoch added nine more external benchmarks covering agentic work, cybersecurity, algorithm engineering, forecasting, and research-level physics. The competitive bar has been raised twice in two weeks, against a model that hasn't yet shipped.
The structural read is that frontier leadership now rotates between three vendors on roughly monthly cadence, which is why model-agnostic deployment architectures, the kind LemonLime and similar platforms have been quietly building toward, have started to look less like a hedge and more like the default. Single-vendor lock-in assumes a stable leader. The benchmarks no longer describe one.
Six days. A preview. A page that says "coming soon."