Gemini 3.5 Pro Delayed a Third Time — Google Is Exploring a Stopgap Gemini 3.6 Flash Instead

What Actually Happened — The Third Slip

Google DeepMind's Gemini 3.5 Pro has encountered another delay, marking its third postponement since the original June 2026 release target. The model has struggled to meet key performance benchmarks, particularly in areas like reliability and real-world application. Specific issues, including frequent hallucinations and inconsistent outputs, have raised concerns about its readiness, especially in a competitive market that includes OpenAI's GPT-5.6.

Early evaluations revealed that older model checkpoints outperformed newer iterations, indicating setbacks in the training process. This is the most damaging technical signal in the delay story. Normally, newer training checkpoints should improve on older ones — that is the entire point of continued training. When the reverse happens, it suggests something in the training dynamics is actively degrading the model's capabilities as it trains further. This is not a tuning problem or a safety review delay. It is a core training instability problem, and it is harder to fix than either of the two previous delay reasons.

The delay timeline tells the story: Gemini 3.5 Pro was promised for June at Google I/O on May 19 — Sundar Pichai told a visibly frustrated crowd of developers, "Give us until next month to get it to you." June came and went. Google pushed to July 17. July 17 came and went. No new date has been announced. Google's official product pages now read "3.5 Pro — coming soon" — the same language used before any target date was circulating.

The Stopgap Plan — Gemini 3.6 Flash and 3.5 Flash Light

In an effort to address these challenges and maintain its market presence, Google DeepMind appears to be exploring interim solutions. Recent registrations for models such as Gemini 3.6 Flash and Gemini 3.5 Flash Light suggest that the company is preparing stopgap releases. These interim models could serve several strategic purposes: buy time to refine and improve Gemini 3.5 Pro, address specific market needs such as enhanced SVG generation and improved natural language processing capabilities, and demonstrate incremental progress to stakeholders and users.

Gemini 3.6 Flash would be a significant strategic admission — Google releasing a Flash-tier model with a higher generation number than a Pro-tier model that has not yet shipped. The naming alone signals that the 3.5 Pro train may have left without the Pro model on it. More consequentially, there is a possibility that Google DeepMind may pivot its focus to future models like Gemini 4.0 Flash, emphasising practical use cases over benchmark performance to regain competitive edge. If Google skips 3.5 Pro and jumps to a 4.0 generation, it would be the clearest signal yet that the 3.5 Pro architecture is fundamentally unworkable.

The Talent Departure Context — Four Senior Researchers Gone

In the week of June 21-27, 2026, four senior Gemini researchers announced they are leaving Google for Anthropic. This follows a broader pattern across 2025 and 2026, where Google has lost key AI researchers to Anthropic, OpenAI, and startups. The timing here is specific: these announcements came while Gemini 3.5 Pro was still in preview and the June GA target had just been missed.

The departures include the AlphaFold core team. Noam Shazeer — co-author of the 2017 Transformer paper and Gemini co-lead — joined OpenAI. John Jumper — Nobel laureate and AlphaFold lead — joined Anthropic. Jonas Adler, who led Google's AI coding effort (the direct competitor to Claude Code), joined Anthropic. Alexander Pritzel, a pretraining specialist and AlphaFold contributor, joined Anthropic. Four departures in one week is notable. Ten departures over a month is a different conversation. The checkpoint regression problem — newer training runs underperforming older ones — may be connected to the loss of researchers who built the pretraining infrastructure these models run on.

What Developers Should Do Right Now

The Competitive Picture Without Gemini 3.5 Pro

Sources: Geeky Gadgets (July 16, 2026) · Windows Forum · TechTimes · Bind AI · StartupFortune · Related: Why Google scrapped the base model → · GPT-5.6 Sol full review → · DeepSeek July 24 migration guide →