Google is facing the heat in the AI race. The company is reportedly months behind schedule on the release of its next flagship AI model – Gemini 3.5 Pro. It is believed that Google was forced to delay the planned launch for Gemini 3.5 Pro after facing a roadblock in development – the model failed to meet internal goals, particularly in coding – an area where OpenAI and Anthropic have largely dominated.
As per a report from Bloomberg, Gemini 3.5 pro was supposed to launch in June after Google CEO Sundar Pichai announced the AI model during Google I/O in May. At the time, Pichai said that the company had already begun using Gemini 3.5 Pro internally.
However, Google engineers are reportedly not happy with the performance of Gemini 3.5. Meanwhile, Anthropic and OpenAI have continued to release newer models, including Claude Fable 5 and GPT-5.6 respectively – two models that are so advanced that the US initially asked the two companies to limit their releases.
Following the report, shares of Alphabet, Google’s parent company, fell by nearly 3 per cent, and at one stage trading down about 4 per cent on Thursday.
Why has Google delayed Gemini 3.5 Pro?
The report claims that Google engineers, AI researchers and managers have grown frustrated with the slow progress of Gemini 3.5 Pro. Internally, employees are worried that the company may lose ground in the AI market as Gemini falls short of the competition particularly in coding.
One person said Google updated the data used to train Gemini late last month in an effort to improve its coding skills, but the results were disappointing.
To give you some context about the importance of coding. In recent months, OpenAI has doubled down on its coding platform Codex in a bid to gain a bigger share of the enterprise market where Anthropic has had the edge.
According to employees cited by the report, Google’s size has also contributed to delays. Multiple layers of stakeholders are involved in model launches at the company, as it tries to apply AI across search, maps, YouTube and other products.
One former employee said getting every department to move in the same direction was like trying to “boil an ocean.” Current and former staff said shifting mandates, duplicated work across teams and competition for resources have made it harder to maintain a clear strategy and to get products into the market quickly.
Google says it is testing Gemini 3.5 Pro
A Google spokesperson told Bloomberg, “We’re shipping quickly across a wide range of models while keeping them highly cost-effective for customers.” The spokesperson added, “We’re currently testing 3.5 Pro, an upgraded Flash model, and other models with partners, and we’re productively engaged with the U.S. government.”
Google has also been in talks with US authorities as scrutiny of advanced AI models increases and companies face questions over testing, safety and national security risks.
This is not to say that Google is completely out of the AI race. The report adds that according to researchers, Gemini’s main strength remains its ability to query Google search data, while Anthropic and OpenAI are viewed as leading on the most powerful general models. Google has pointed to other strengths, including work across text, images and video, as well as progress on world models that mimic physical environments.
After ChatGPT’s launch in late 2022 raised fears that Google’s search business could be disrupted, the company declared a “code red” to cut through bureaucracy and internal competition. Google co-founder Sergey Brin and others had reportedly pushed for faster progress in AI coding, but those efforts were slowed by competing factions.
Google Cloud, Google DeepMind and the Android team are all building AI coding tools, with some consumer product teams also involved.
It is believed that Google employees faced early restrictions on using Gemini to write or analyse software because of concerns that proprietary code could leak into training data. Those policies have since been relaxed, but former staff said they limited early experimentation.
Google recently said that 75 per cent of code at the company was now generated by AI and reviewed before reaching production. The tech giant has also streamlined some coding tools under Google Antigravity. Employees are now expected to use AI to generate code, but it is said that they do face capacity constraints because of competition for computing power.
Frustration over Google’s position in the AI race has also contributed to departures to Anthropic and other labs. Access inside Google to Anthropic’s Claude has meanwhile been restricted to teams working on cutting-edge research and other high-priority projects.
