According to informed sources, the release of Alphabet's flagship AI model, Gemini 3.5 Pro, has been postponed by several months from its original schedule. The company is dedicating additional time to enhance the model's capabilities, with a particular focus on its programming proficiency.
Ten current and former employees have indicated that the delay in launching Gemini 3.5 Pro has caused dissatisfaction among engineers, AI researchers, and management at the company. There is concern that as competitors like Anthropic and OpenAI release models that outperform Gemini, Alphabet could lose its competitive edge in the market.
These sources note that the release process is slowed by the need for approvals from multiple stakeholders and the integration of AI capabilities into products like Search, Maps, and YouTube. The individuals spoke on condition of anonymity due to the sensitive nature of the internal information.
Both OpenAI and Meta Platforms Inc. have recently unveiled new models that demonstrate superior programming capabilities compared to Alphabet's offerings. One source mentioned that Alphabet updated the training data for Gemini at the end of last month in an attempt to boost its coding skills, but the results were disappointing. Shares of Alphabet (NASDAQ: GOOGL) fell as much as 3.2% on Thursday.
In a statement, an Alphabet spokesperson said, "We are accelerating the release of multiple model classes while maintaining high cost efficiency." The spokesperson added that the company is communicating with the U.S. government regarding Gemini's capabilities and AI safety standards.
The statement further noted, "We are currently testing 3.5 Pro, an upgraded Flash model, and other models with partners, and actively collaborating with the U.S. government on model testing and broader regulatory framework discussions."
Earlier this year, Anthropic temporarily withdrew its latest model under U.S. government pressure after internal testing revealed potentially dangerous cybersecurity capabilities. OpenAI has also voluntarily limited and phased the rollout of its newest AI models due to national security concerns and pressure from government officials.
Alphabet's extensive product ecosystem provides an entry point for general users to access generative AI and can leverage data from these products to improve model performance. However, a former employee stated that aligning all business units is as challenging as "trying to boil the entire ocean."
Current and former employees said maintaining a unified strategy becomes more difficult when company priorities shift or multiple departments work redundantly on the same tasks. This also makes it harder for individual projects to secure sufficient resources and market attention.
Informed sources revealed that teams within Google Cloud, Google DeepMind, and Android are all developing AI programming tools for developers, with some consumer product teams also involved.
Former employees indicated that some engineers within the company hold a more conservative view, believing all critical code should be human-written to meet internal standards. They also noted that when AI tools were first introduced, employees were restricted from using Gemini to write or analyze software code due to concerns about proprietary code leaking into model training data. While these rules have since been relaxed, they initially limited engineers' opportunities to experiment with AI development.
Alphabet stated that it recently disclosed at a Cloud event that approximately 75% of code is now AI-generated and goes into production after human review, a process that aligns with the company's standards.
Today, Alphabet engineers are encouraged to use AI for code generation. However, employees report that in practice, they often face capacity limits due to competition for the company's internal computing resources.
AI researchers point out that Gemini's primary advantage is its ability to leverage Alphabet's search data, while Anthropic and OpenAI currently lead in building the most powerful AI models.
Former Alphabet employees said some researchers, disappointed with the company's position in the AI race, have left for leading AI labs like Anthropic.
Currently, only select teams within Alphabet have access to Anthropic's Claude model. The company has restricted use of the tool to teams working on cutting-edge research and other high-priority projects.
While waiting for Gemini 3.5 Pro, customer feedback on the existing Gemini 3.5 Flash has been mixed. Rodrigo Davies, a product manager at design platform Figma, stated that the company recently integrated 3.5 Flash into its new "Figma Agent" AI assistant to help designers generate and iterate on ideas. For Figma, the model strikes an ideal balance between speed and quality.
In contrast, Freddy Vega, founder and CEO of the Latin American edtech platform Platzi, described 3.5 Flash as being in an awkward position. He noted it is priced higher than Alphabet's previous 3.1 Flash model but is slower, and its capabilities remain significantly inferior to competitors' high-end models.
He said his team has now switched to using a mid-tier model from Anthropic to balance reasoning ability and response speed.
Comments