Google delays Gemini 3.5 Pro as it struggles to meet internal performance targets: Report

Google has delayed the release of Gemini 3.5 Pro, its next flagship AI model, by several months as it works to improve the model's overall performance, particularly its coding capabilities, according to people familiar with the matter.

The delay has reportedly frustrated Google engineers, AI researchers, and managers, who fear the company could lose ground to rivals such as Anthropic and OpenAI, whose latest models are seen as outperforming Gemini in several areas. Sources said Google's extensive internal review process and the need to integrate AI across products such as Search, Maps, and YouTube have also contributed to the slower rollout.

Engineers and AI researchers are reportedly concerned that Google risks losing its competitive edge as rivals Anthropic and OpenAI continue to release models that outperform Gemini in several areas, according to 10 current and former employees.

Sources said Google's complex internal structure has also contributed to delays. The company has multiple layers of stakeholders involved in preparing AI models for release while simultaneously integrating AI across a broad portfolio of products, including Search, Maps, and YouTube. Google reportedly updated the data used to train Gemini late last month in an effort to improve its capabilities, particularly in coding, but the results fell short of expectations.

More from Tech

Current and former employees also said Google has struggled to maintain a cohesive AI strategy. Shifting priorities and overlapping responsibilities across teams have complicated execution, while the fragmented structure has made it more difficult for individual products to secure the resources needed to succeed and gain traction in the market.

Google 'Code Red'

Following the launch of ChatGPT in late 2022, concerns that Google's search business could be disrupted prompted the company to declare a "Code Red." According to a former employee, the move was intended to cut through the bureaucracy and internal competition that often slowed decision-making.

Reports suggest that while co-founder Sergey Brin pushed Google to accelerate its AI efforts, progress was slowed by competing priorities across different teams. Some engineers also maintained a traditional view that critical code should continue to be written by humans to meet Google's engineering standards, former employees said. In addition, employees were initially restricted from using Gemini for certain coding tasks over concerns that proprietary code could be incorporated into the model's training data.

Google has since streamlined several of its internal AI coding tools by consolidating them under Google Antigravity, a platform designed to provide shared infrastructure for data, memory, and AI safety protocols.

A dedicated Deep Mind team at work

Google has been reorganising its AI efforts internally, with Chief AI Architect Koray Kavukcuoglu working to unify the company's coding tools and a dedicated DeepMind team focusing on AI coding. However, employees reportedly face computing capacity constraints when using AI to generate code, limiting productivity.

Google has also restricted access to Anthropic's Claude internally, limiting its use to teams working on cutting-edge research and other high-priority projects.

Early customer feedback on Gemini 3.5 Flash has been mixed. While Figma praised the model for balancing speed and quality, education platform Platzi said it is slower and more expensive than Gemini 3.1 Flash while remaining less capable than competing premium models. The company has since shifted many workloads to Anthropic's Claude 3.5 Sonnet for tasks requiring stronger reasoning.