(Bloomberg) — Alphabet Inc.’s Google is months behind schedule on delivering Gemini 3.5 Pro, its most powerful flagship AI model, because the company has been taking time to try to improve its capabilities, particularly in coding, according to people familiar with the matter.
Most Read from Bloomberg
The delay has been a source of frustration for Google engineers, AI researchers and managers, many of whom are concerned the company risks losing an edge in the market as rivals Anthropic and OpenAI produce models that exceed Gemini’s capabilities, according to 10 current and former employees. Google has multiple layers of stakeholders involved in preparing models for release, working to weave AI across a vast product portfolio, including search, maps and YouTube, which can cause delays, said the people, who declined to be named discussing internal concerns.
Both OpenAI and Meta Platforms Inc. recently released new models that further outpace Google’s current offerings in AI for writing code. Late last month, Google updated the data being used to train Gemini in an attempt to improve these skills, but the results were disappointing, one of the people said.
“We’re shipping quickly across a wide range of models while keeping them highly cost-effective for customers,” a Google spokesperson said in a statement. Google has also been in talks with the US government, which has been increasingly monitoring AI companies’ most advanced models, about its capabilities, as well as the standards that should be applied to the industry for safety. “We’re currently testing 3.5 Pro, an upgraded Flash model, and other models with partners, and we’re productively engaged with the US government on model testing and broader frameworks.”
Early this year, Anthropic faced whiplash from the US after internal testing flagged dangerous cybersecurity capabilities in its latest models, forcing the startup to temporarily pull them. OpenAI has voluntarily limited and staggered the release of its newest AI model after facing national security concerns and significant pressure from the Trump administration.
Google’s popular products are a gateway to generative AI for everyday people, and can yield data that makes their answers smarter. But encouraging leadership of every department to move in the same direction is like trying to boil an ocean, one ex-employee said. When mandates shift or efforts end up duplicated in multiple departments, it gets even more difficult to maintain a cohesive strategy, current and former employees said. It’s also a challenge for any one offering to get the resources it would need to succeed, and to gain traction in the market, they said.
After the launch of ChatGPT in late 2022 sparked concerns that Google’s search engine would become obsolete, the company declared a “code red” — a useful tactic for cutting through the layers of bureaucracy and internal competition that often slow Google’s product efforts. But now, racing in AI is the normal state of the company, one employee said.
Google co-founder Sergey Brin and others were advocating for Google to move faster to seize opportunities in AI coding, but their efforts were slowed by competing factions within the company, two former employees said. Cloud computing unit Google Cloud, research lab Google DeepMind and the team behind the Android operating system are all building AI coding tools for developers, with involvement from some consumer product teams, too, people familiar with the work said.
Efforts to win at coding have also been up against some engineers at Google with a more purist stance, who believe that all important code should be human-written to adhere to Google standards, ex-employees said. Early in the rollout of the technology, employees also faced restrictions on using Gemini to write or analyze software over concerns that proprietary code could leak into the AI model’s training data, they said. Those policies, which have since been relaxed, limited opportunities for engineers to experiment with AI development.
Google said it announced at its most recent Cloud conference that 75% of code at the company is now generated by AI, meaning it’s being reviewed and surviving to production — and meeting Google’s standards. The company also said it streamlined some of its coding tools across products, mostly consolidating them under Google Antigravity, which provides the scaffolding for data, memory and safety protocols that the AI needs to interact with operating systems and applications.
Google is taking steps to reduce internal confusion. Chief AI Architect Koray Kavukcuoglu is working with Google’s main engineering team to unite the company’s internal artificial intelligence coding tools. And earlier this year, the company formed a team within DeepMind to tackle AI coding, led by research engineer Sebastian Borgeaud, Bloomberg has reported.
Engineers within Google are now expected to use AI to generate code. But when they try to use AI, they often hit capacity constraints due to competition for computing power within Google.
AI researchers say that Gemini’s strongest selling point is querying Google search data, while Anthropic and OpenAI have taken the lead on building the most powerful models. Google says it has other strengths in AI, like the ability to work with various types of input like images or videos, and progress in AI world models, which can mimic physical environments.
Some researchers’ frustration with Google’s position in the AI race has contributed to a wave of departures to Anthropic and other top labs, according to former employees.
Only some teams are allowed to use Anthropic’s Claude. Access became restricted to teams doing cutting-edge research and other high-priority projects. In addition, only some teams are allowed to use Anthropic’s Claude — Google restricted the rival tool to teams doing cutting-edge research and other high-priority projects.
While they wait for the 3.5 Pro version of Gemini, Google customers have had a mixed experience with Gemini 3.5 Flash. Rodrigo Davies, a product manager at design platform Figma, said the company recently added 3.5 Flash to its newly launched “Figma agent,” an AI assistant that helps designers generate and iterate on ideas. For Figma, the model hit a sweet spot of speed and quality.
But Freddy Vega, chief executive officer and founder of Platzi, a Latin American education-technology platform, said 3.5 Flash occupies an awkward middle ground: It is more expensive than Google’s previous 3.1 Flash model, yet slower, and it remains far less capable than premium offerings from competitors. He said the model often struggles with structured data.
For tasks requiring a balance of speed and reasoning, his team has shifted away from Google to Anthropic’s mid-tier model, Claude 3.5 Sonnet.
–With assistance from Shirin Ghaffary.
Most Read from Bloomberg Businessweek
©2026 Bloomberg L.P.
