# Gemini 3.5 Pro Keeps Missing: Google’s Third Launch Failure > Source: > Published: 2026-07-22 19:16:30+00:00 Gemini 3.5 Pro, the flagship model Sundar Pichai personally promised at Google I/O on May 19 — “give us until next month” — has now missed its public launch deadline three times. As of today, July 22, the model sits in limited enterprise preview with no confirmed general availability date. This week, Bloomberg and 9to5Google reported that a late-June training data update specifically targeting the model’s coding failures produced disappointing results. The fix made things worse. ## What Broke in Gemini 3.5 Pro — and Why Google Can’t Fix It Gemini 3.5 Pro’s core problem is coding performance. Google’s own engineers reportedly identified structural failures in recursive tool-calling and SVG generation — the kind of tasks that define enterprise-grade agentic workflows. When that became clear after the June miss, Google updated training data in late June specifically to address coding. The result: disappointing, per 9to5Google’s reporting on [the delays published July 16](https://9to5google.com/2026/07/16/gemini-3-5-pro-delays/). This matters because coding competence is the entire value proposition for Gemini 3.5 Pro. The 2 million token context window — twice Claude Opus 4.8’s limit — is genuinely compelling for full-codebase ingestion. However, a large context window paired with unreliable code generation is a niche product, not a flagship. Google hasn’t solved the fundamental issue, and a second patch hasn’t shipped. ## Google Paid $2.7 Billion to Keep the Transformer Co-Inventor — Then Lost Him to OpenAI On June 18 — the same week the June GA target was missed — Noam Shazeer resigned from Google to join OpenAI as Lead for Architecture Research. Shazeer co-invented the Transformer architecture in 2017, co-founded Character.AI, and returned to Google in August 2024 when [Google paid $2.7 billion for Character.AI’s assets and key personnel](https://www.cnbc.com/2026/06/18/google-gemini-co-lead-noam-shazeer-leaves-for-openai.html). He lasted less than two years. Sam Altman welcomed him on X: “one of the people I have most wanted to work with since the very beginning of OpenAI.” That same week, three more senior DeepMind researchers joined Anthropic: Nobel Chemistry laureate John Jumper (AlphaFold2), Jonas Adler, and Alexander Pritzel. Four senior researchers gone in seven days. Google’s public response was a statement about shipping quickly across “a wide range of models while keeping them highly cost-effective” — which describes the Flash tier. It is not a response to losing your flagship model’s co-lead and your Nobel Prize winner in the same week. Related:[Gemini 3.6 Flash: 17% Fewer Tokens, Live in Copilot] ## Three Options That Aren’t Waiting Developers who planned around Gemini 3.5 Pro for H2 2026 need an interim strategy. Three concrete paths exist. First, [Gemini 3.6 Flash launched July 21](https://github.blog/changelog/2026-07-21-gemini-3-6-flash-is-now-available-in-github-copilot/) with meaningful improvements: 17% fewer output tokens, a DeepSWE coding score of 49% (up from 37% for 3.5 Flash), and a knowledge cutoff advanced to March 2026. At $1.50 input / $7.50 output per million tokens, it handles routine agentic workloads without the Pro-tier price or the wait. Second, if 2 million token context is a genuine production requirement, apply for Vertex AI enterprise preview now — the waitlist exists but access is being granted. Third, and this is the practical option most teams won’t take but should, set a decision deadline. If Gemini 3.5 Pro isn’t generally available by August 15, commit to an alternative for the rest of 2026. Claude Opus 4.8 handles 1 million tokens reliably today. GPT-5.6 Sol now has Shazeer working on its architecture. [Both are available](https://www.axios.com/2026/06/18/noam-shazeer-google-openai-characterai). ## One More Signal Developers Should Notice On July 21, the same day Google shipped Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber, the company announced Gemini 4 pretraining has begun — described as the company’s “most ambitious run yet.” Gemini 3.5 Pro is still unshipped. Google is announcing its next generation while its current flagship misses a third deadline. For developers making architecture decisions now, this signals that Gemini 3.5 Pro may eventually ship in a diminished role — a gap-filler between the existing Flash tier and whatever Gemini 4 becomes. If that is the trajectory, waiting for it is even harder to justify. ## Key Takeaways - Gemini 3.5 Pro has missed its launch window three times; a late-June coding fix produced disappointing results and no new GA date has been confirmed - Google lost Gemini co-lead Noam Shazeer to OpenAI and three senior researchers — including Nobel laureate John Jumper — to Anthropic in the same week the June target was missed - Gemini 3.6 Flash, released July 21, is the practical alternative for most agentic workloads today: improved coding scores, 17% token reduction, available now - Set a concrete decision deadline — if Gemini 3.5 Pro is not generally available by mid-August, commit to an available model for the rest of 2026