Google has postponed the launch of Gemini 3.5 Pro, its most advanced AI model, after recent internal tests failed to meet expectations — particularly in programming tasks. The decision adds pressure to Alphabet’s AI strategy at a moment when OpenAI and Anthropic are accelerating their release cadence.
Prioritizing Improvements Over Deadline
Gemini 3.5 Pro was initially scheduled for release in June following Sundar Pichai’s announcement at Google I/O. Instead, Google extended development to improve performance before making the model public. According to people familiar with the project, the company recently updated the data used during training to strengthen code generation — and the results so far have not met the established internal goals.
Competition Raises the Stakes
The delay lands amid intense competition. OpenAI recently unveiled GPT-5.6, while Anthropic continues developing its models after overcoming temporary restrictions from US export controls. Within Google, some engineers, researchers, and managers are reportedly concerned the company may be losing ground to rivals in crucial areas like reasoning and programming.
The Integration Factor
Gemini launches also require coordinating numerous teams because the model must be integrated into Google Search, Maps, and YouTube. That strategy aims for consistent performance across Alphabet’s entire product ecosystem, but it increases the complexity of every update — and, sources say, helps explain the longer timeframes for new versions.
Alphabet’s Response
After the news broke, Alphabet shares dropped nearly 3%. The company says it continues to test Gemini 3.5 Pro, alongside an improved Flash model, with various strategic partners. A Google spokesperson stated the company keeps launching new models while working toward efficient, cost-effective solutions, and confirmed ongoing collaborations with the US government on responsible AI development.
What This Means
The delay reflects the rising technical bar for frontier models: reasoning quality, programming performance, and operating costs now decide who wins enterprise deals. Shipping late with stronger code generation may cost Google a quarter of headlines — but shipping a weak programmer into a market defined by agentic coding could cost far more.