Google (GOOGL.US) delays the launch of Gemini 3.5 Pro large language model as code generation capability falls short of goals

According to Zhitong Finance APP, Alphabet (GOOGL.US, GOOG.US) shares fell 4% on Thursday due to news of a delay in the release of its flagship artificial intelligence model. Media reports indicate that, as the model has not yet met internal expectations, Google has decided to postpone the launch of the Gemini 3.5 Pro large language model, pushing its release date back by several months from the original schedule.

According to sources familiar with the matter, the development progress of Gemini 3.5 Pro has lagged behind expectations, mainly because Google is still working to optimize its performance. In particular, its code generation capabilities have not reached the company's internal targets, and this area has become one of the most fiercely contested tracks among global AI companies.

The reports noted that as Google slows its product release pace, OpenAI and Meta Platforms (META.US) have recently launched new generations of AI models, whose code generation capabilities generally outperform Google’s current products, further intensifying market competition.

Google first announced Gemini 3.5 Pro at the Google I/O developer conference in May this year and stated that the model had already begun internal use within the company, with plans to make it available to more users the following month. However, the latest update shows that this plan has been postponed.

In response to related reports, an Alphabet spokesperson stated that the company continues to rapidly launch multiple AI models while maintaining high cost efficiency. The spokesperson said the company is currently testing Gemini 3.5 Pro, the upgraded Flash model, and other AI models with partners, and is actively cooperating with the U.S. government.

In recent years, code generation has become one of the most important application scenarios for generative AI and a key point of competition among major AI model providers. Companies such as OpenAI and Anthropic continue to launch code generation models aimed at developers, with some open-source models available for developers to use free of charge.

Just last week, Meta released the new-generation large language model Muse Spark 1.1. Meta Chief AI Officer Alexandr Wang said this is the company’s strongest model to date in terms of AI agent and code generation performance.

Likewise last week, OpenAI launched the GPT-5.6 Sol model. OpenAI CEO Sam Altman stated that the new model’s token utilization efficiency in AI agent code generation tasks has improved by 54%, further enhancing the model’s competitive edge in terms of performance and cost.

Analysts believe that as AI model competition gradually shifts from parameter scale to practical application capabilities and cost efficiency, code generation capability has become an important metric for evaluating the competitiveness of large language models. The delayed release of Gemini 3.5 Pro also reflects Google's desire to only release the model to the market once its performance is mature, in order to address the increasingly intense industry competition.