Sources with insider knowledge reveal that Google has decided to delay the release of its premier AI model, Gemini 3.5 Pro, by several months from its initial timetable. The postponement aims to further elevate the model’s abilities, especially in the realm of programming. This setback has triggered discontent among Google’s internal engineering teams, AI researchers, and management. They worry that Google might forfeit its competitive advantage in the market, given that Anthropic and OpenAI are persistently unveiling increasingly potent models.

Originally, Google had earmarked June 2026 for the model’s debut but subsequently rescheduled it to July. The motive behind this shift was to polish the model based on insights gleaned from preliminary testing. Emphasis was placed on enhancing its performance in long-text tasks and agent-driving functionalities, as well as rectifying problems like excessive token usage. Internal evaluations indicate that Gemini 3.5 Pro’s programming capabilities are nearly on par with those of GPT-5.5. Notable strides have been made in cross-modal image-text comprehension and code generation tasks. Nevertheless, despite substantial efforts invested in refinement, the test outcomes still fell below the anticipated benchmarks, exacerbating internal apprehensions.