Google Had the Perfect Window. Gemini 3.5 Pro Is Still in Preview.

In the second week of June 2026, the competitive landscape of frontier AI did something it hadn't done before: both of Google's primary competitors stumbled simultaneously. Claude Fable 5 was pulled offline by a US Commerce Department export control directive on June 12 — the first time a frontier AI model had been recalled by government order. GPT-5.6, OpenAI's next model, slipped its June launch, with prediction markets collapsing from 83% to 18% odds before month-end. For a window of weeks, Google had a clear lane: the only company with a capable frontier model in GA, positioned to absorb enterprise customers who needed something Fable 5 or GPT-5.6 couldn't deliver.

Gemini 3.5 Pro — announced at Google I/O in May, featuring a 2M token context window and Deep Think reasoning — has not shipped publicly. As of June 24, it remains in limited Vertex AI enterprise preview. Secondary aggregators reported it as "confirmed for GA June 23-30." Pichai's actual words at Google I/O were "give us until next month." Six days remain in June.

What the model actually is

Gemini 3.5 Pro's confirmed features, as verified through enterprise preview, are substantive. The 2M token context window is real — developers in preview have documented loading full 2,000-file TypeScript monorepos and multi-year legal document sets without chunking. Deep Think reasoning mode is confirmed as an API thinkingConfig parameter, gated to the $250/month Ultra subscription tier. Pricing structure circulating from preview users: $15/M input, $60/M output for Pro.

What doesn't exist: public benchmarks. No model card, no independent evaluation, no official Google performance data has been published for Gemini 3.5 Pro. Coverage citing benchmark scores for the Pro model is using Gemini 3.5 Flash or Gemini 3.1 Pro numbers — a different model. FrankX.AI's pre-GA analysis is explicit: "If you see a '94% on benchmark X' claim for 3.5 Pro right now, it is either extrapolated from Gemini 3.1 Pro / 3.5 Flash or invented." Gemini 3 Deep Think achieved 84.6% on ARC-AGI-2; GPT-5.5 leads at 85.0%. Whether 3.5 Pro extends that lead is unknown.

The context quality problem

Research consensus on context windows is worth stating plainly because it rarely makes it into coverage of context window announcements: context windows grew 62× in two years; practical utility grew approximately 3×. Most frontier models — including Gemini 3.1 Pro, which preceded 3.5 — lose coherence and retrieval accuracy at 60–70% of their advertised maximum token count. Gemini 3.1 Pro degraded significantly past 500K tokens in documented developer testing despite higher advertised limits.

Whether Gemini 3.5 Pro maintains useful quality at 2M tokens is an open question. The developer use cases circulating from enterprise preview — full codebase analysis, multi-decade legal review — are exactly the scenarios where quality at the tail end of the context window determines whether the product is useful or merely impressive in a demo. "Largest context window" is a marketing number until stress tests run.

The competitive logic still holds even with this caveat. If Gemini 3.5 Pro degrades at 1.4M instead of 2M, it still has a meaningful context advantage over GPT-5.5's 1M. Whether that advantage translates into enterprise conversions depends on whether the quality threshold falls within the actual use case requirements.

The competitive window, precisely stated

The strategic situation is: Anthropic's primary frontier model is suspended by government order. The suspension has now lasted 12 days with no confirmed restoration date. Enterprise customers who needed Fable 5 for long-context, reasoning-heavy workloads are either on Opus 4.8 (200K context ceiling) or shopping for alternatives. OpenAI's next context upgrade is in July. Gemini 3.5 Flash is already GA at 1M context but sits a tier below Pro in capability.

Google has not shipped the product built to fill this window. Whether this is a readiness problem, a quality assurance issue, or standard enterprise rollout caution is not publicly knowable from outside. Google's track record on announced windows is mixed: Gemini 3.1 Pro launched on schedule; the broader Gemini 3.5 rollout has shipped Flash before Pro as expected. A late-June GA is structurally consistent with prior behavior.

The window closes at the end of June — when GPT-5.6 presumably arrives, when prediction markets suggest the Fable 5 suspension may resolve, and when the competitive configuration that made mid-June unusual reverts toward a more normal frontier state.

Six days

If Gemini 3.5 Pro ships before June 30 with context quality that holds in enterprise stress testing, Google will have taken one of the cleanest competitive openings available to it in 2026. The first independent benchmark data will establish whether the model closes the gap with GPT-5.5 on reasoning or opens a new one on context.

If it doesn't ship before June 30, the competitive window closes without Google filling it. GPT-5.6 arrives, the frontier reorganizes, and the missed opportunity becomes a footnote rather than a narrative. The model will still ship; the timing will just matter less.

I don't know which outcome lands. Neither does anyone who isn't inside Google's DeepMind infrastructure team. The prediction markets put "released by June 30" odds at roughly 50–55% as of mid-June. That number is not "confirmed." It is a coin flip with better-informed bettors.