Google Gemini 3.5 Pro Launch Delayed as Tech Falls Short of Internal Goals: Report

Google is months behind schedule in delivering the Gemini 3.5 Pro, its most powerful flagship AI model, as the company is taking its time to work on improving its capabilities, particularly in coding, according to people familiar with the matter.

The delay has been a source of frustration for Google engineers, AI researchers and executives, according to 10 current and former employees; Many of them are concerned that the company risks losing its advantage in the market as rivals Anthropic and OpenAI produce models that exceed Gemini’s capabilities. Google has multiple layers of stakeholders involved in preparing models for release, working to weave AI across a broad portfolio of products including search, maps and YouTube, which could cause delays, said the people, who declined to be named discussing internal concerns.

Both OpenAI and Meta Platforms have recently released new models that surpass Google’s current AI offerings for writing code. Late last month, Google updated the data used to train Gemini to improve these skills, but one of the people said the results were disappointing. Shares lost as much as 3.2% on Thursday.

“We ship quickly on a wide range of models while keeping it extremely affordable for customers,” a Google spokesperson said in a statement. Google was expected to release 3.5 Pro at its developer conference in May. Google is also in talks with the US government, which is increasingly monitoring the most advanced models of artificial intelligence companies, about the capabilities of these models, as well as the standards that should be applied to the industry in terms of security. “We are currently testing the 3.5 Pro, an upgraded Flash model, and other models with our partners, and we are working productively with the US government on model testing and larger frameworks.”

Earlier this year, Anthropic faced harsh US backlash after internal testing flagged dangerous cybersecurity features in its latest models, forcing the startup to temporarily pull those features. OpenAI voluntarily limited and staggered the release of its newest AI model after facing national security concerns and severe pressure from the Trump administration.

Google’s popular products are a gateway to productive AI for ordinary people and can provide data that makes their responses smarter. But encouraging each department’s leadership to move in the same direction is like trying to boil the ocean, one former employee said. Current and former employees said it becomes even more difficult to maintain a coherent strategy when reshuffles or efforts result in duplication of efforts across multiple departments. They said it’s also a challenge for anyone proposing to acquire the resources they’ll need to succeed and gain market traction.

After ChatGPT’s launch in late 2022 raised concerns that Google’s search engine would become obsolete, the company announced “code red,” a useful tactic to bypass the layers of bureaucracy and internal competition that often slow Google’s product efforts. However, one employee said that competing with artificial intelligence is now the normal state of the company.

Google co-founder Sergey Brin and others had advocated for Google to move faster to seize opportunities in AI coding, but their efforts were slowed by rival groups within the company, two former employees said. Cloud computing unit Google Cloud, research laboratory Google DeepMind and the team behind the Android operating system are developing artificial intelligence coding tools for developers, with the participation of some consumer product teams, people familiar with the work said.

His efforts to win at coding also ran afoul of some engineers at Google with a purer stance, believing that all important code should be written by humans to comply with Google standards, former employees said. They said early in the technology’s rollout, employees faced restrictions on using Gemini to write or analyze software due to concerns that custom code could leak into the AI ​​model’s training data. These policies, which have since been relaxed, limited opportunities for engineers to experiment with developing artificial intelligence.

At its latest Cloud conference, Google announced that 75% of the code at the company is now created by AI, meaning the code has been reviewed and survived to production and meets Google’s standards. The company also said it is streamlining some coding tools across products, mostly unifying them under Google Antigravity, which provides scaffolding for the data, memory and security protocols that AI needs to interact with operating systems and applications.

Google is taking steps to reduce internal confusion. Chief AI Architect Koray Kavukçuoğlu is working with Google’s core engineering team to unify the company’s internal AI coding tools. According to Bloomberg, earlier this year, the company established a team within DeepMind, led by research engineer Sebastian Borgeaud, to tackle artificial intelligence coding.

Engineers at Google are now expected to use AI to create code. But when they try to use AI, they often face capacity constraints due to competition for computing power from Google.

Artificial intelligence researchers say that Gemini’s most powerful feature is querying Google search data, while Anthropic and OpenAI take the lead in creating the most powerful models. Google says it has other strengths in AI, such as the ability to work with various types of input, such as images or videos, and progress in AI world models that can mimic physical environments.

Frustration by some researchers with Google’s position in the AI ​​race has contributed to a wave of departures to Antropik and other top labs, according to former employees.

Only some teams are allowed to use Anthropic’s Claude. Access became limited to teams doing cutting-edge research and other high-priority projects.

While waiting for the 3.5 Pro version of Gemini, Google customers had a mixed experience with Gemini 3.5 Flash. Rodrigo Davies, product manager at design platform Figma, said the company recently added 3.5 Flash to its “Figma agent,” an AI assistant that helps designers generate and iterate ideas. For Figma, the model has reached a perfect point in terms of speed and quality.

But Freddy Vega, founder and CEO of Latin American edtech platform Platzi, said 3.5 Flash sits in an awkward middle ground: It’s more expensive than Google’s previous 3.1 Flash model, but still slower and far less capable than competitors’ premium offerings. He said the model often had problems with structured data.

For tasks that required a balance of speed and reasoning, his team switched from Google to one of Anthropic’s midrange models.