The search giant pushed its flagship AI model from June to July 2026 as competitive pressure from OpenAI and xAI mounts
Google is months behind schedule on the release of Gemini 3.5 Pro as the company works to improve the coding capabilities of its most powerful flagship AI model.
The delay has frustrated engineers, researchers and managers who are concerned that Google could lose ground to Anthropic and OpenAI, according to current and former employees familiar with the matter.
Google was widely expected to unveil Gemini 3.5 Pro during its developer conference in May. The company updated the model’s training data late last month in an effort to improve coding performance, but the results were disappointing.
Alphabet shares fell as much as 3.2% on Thursday following the report.
The company said it is testing Gemini 3.5 Pro, an upgraded Flash model and other systems with partners. Google is also discussing the model’s capabilities and testing standards with the US government.
The delay comes as OpenAI and Meta release models that outperform Google’s current offerings on coding tasks. Researchers inside Google increasingly view access to Search data as Gemini’s strongest advantage, while rivals lead in overall model capabilities.
Google’s size has complicated its response. Google Cloud, DeepMind, Android and several consumer product teams have developed separate AI coding tools, creating overlapping efforts and competition for resources.
Google cofounder Sergey Brin and other executives have pushed the company to move faster in AI coding, but internal divisions have slowed progress.
Some engineers also initially resisted using AI generated code, arguing that important software should remain human written to meet Google’s standards. Earlier policies restricted employees from using Gemini to analyze proprietary code because of concerns that internal information could enter its training data.
Those restrictions have since been relaxed. Google now expects engineers to use AI for coding and said 75% of its code is generated by AI, reviewed by employees and ultimately used in production.
The company has consolidated several coding products under Google Antigravity, which provides data, memory and safety infrastructure for AI systems interacting with operating systems and applications.
Chief AI Architect Koray Kavukcuoglu is working with Google’s engineering organization to unify internal AI coding tools. DeepMind has also formed a dedicated coding team led by research engineer Sebastian Borgeaud.
Employees still face capacity limits when using AI tools because teams compete for computing resources. Frustration over Google’s position has contributed to researchers leaving for Anthropic and other laboratories.
Google customers have reported mixed results from Gemini 3.5 Flash while waiting for the Pro model.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our