Google employees are unsure of Gemini 4 Argon's real-world performance
What's the story
Google's latest AI model, Gemini 4 Argon, is facing internal skepticism over its performance in key areas like coding. Despite performing well on industry-standard benchmarks, employees have found it underwhelming in real-world applications. The model has struggled with specific coding tasks, sources familiar with the matter told Bloomberg.. In response to the internal skepticism, Google has denied claims of underperformance in areas like coding.
Company stance
Google denies underperformance claims
Google referred Bloomberg to comments made by Koray Kavukcuoglu, head of Google DeepMind, who expressed optimism about the model's performance.
"I have the utmost trust in the team," Kavukcuoglu said at a conference hosted by tech news site The Information.
In my mind, it's a certainty that we are always gonna be at the frontier," he added.
Internal debate
Divide among employees
There is a divide among Google employees over the performance of Gemini 4 Argon.
Some believe that Anthropic's Fable and OpenAI's Astra models are advancing faster than Gemini, while others think the upcoming version has caught up with leading AI labs.
A Google employee familiar with model development said there is "large consensus" internally at the company that Gemini 4 Argon is at the frontier.
Product integration
Success of Gemini 4 critical for Google
Google needs Gemini 4 to succeed as it powers nearly all of its products, from AI answers on Search to Maps, Gmail, and Chrome.
Each of these products has over a billion users, giving Google an edge over some competitors.
However, with OpenAI and Anthropic moving beyond model sales to building their own products like coding agents, failing to deliver a cutting-edge model could give them an advantage in convincing consumers and businesses about their platforms.