Press reports have revealed mounting internal skepticism among Alphabet employees regarding the efficiency of the Gemini 4 model in performing real-world programming tasks, despite posting impressive benchmark results. This gap is raising widespread concern given that billions of users rely on Google products powered by this model.
- A concerning gap between theoretical tests and daily practical tasks
- Concerns over a widening competitive gap with OpenAI and Anthropic
- Google management denies setback amid a slight dip in stock
- Frequently asked questions
A concerning gap between theoretical tests and daily practical tasks
Alphabet is facing a rising wave of internal skepticism and reservations among its engineers and employees regarding the actual performance of its anticipated flagship artificial intelligence model, Gemini 4. Company workers have reported that the system encounters tangible difficulties and hurdles when tasked with complex real-world programming duties, despite achieving high scores in established industry standard evaluations, according to an investigative report published by Bloomberg on Wednesday.
This wide gap between standard benchmark scores and applied performance in real-world work environments is sparking deep concern within the halls of the tech giant as it puts the final touches on launching its most crucial model. This issue gains extreme sensitivity given that Gemini models now form the operational foundation for all core services provided by Google—ranging from generative answers in the search engine to Maps, Gmail, and the Chrome browser—services each relied upon by over a billion users worldwide.
Concerns over a widening competitive gap with OpenAI and Anthropic
Individuals familiar with the development path told Bloomberg that despite the encouraging figures Gemini 4 achieves in common benchmark metrics, its effectiveness drops noticeably when employees apply it to real projects and code, particularly in sensitive programming tasks that represent one of the most commercially profitable and valuable applications in the large language model market.
A segment of Google employees believes that main competitors—specifically Anthropic with its internal model known as “Fable,” and OpenAI with its “Astra” model—are achieving rapid development leaps that outpace the pace at which Google is progressing with Gemini. The report recalled that Google had previously made plans to release an advanced version named “Gemini 3.5 Pro” last June, but was forced to cancel the project and abandon it later.
Google management denies setback amid a slight dip in stock
In contrast, Google rushed to deny and refute allegations describing Gemini 4 as having poor performance, relying on public statements made by Koray Kavukcuoglu, head of Google DeepMind’s research lab, who emphasized his encouragement and complete satisfaction with the model’s efficiency and evolution. A responsible Google employee closely familiar with training operations told Bloomberg that there is a “broad internal consensus” that Gemini 4 stands at the forefront of global frontier models, strongly denying that the model faces any exceptional difficulties in executing real-world code.
Other engineers shared this optimistic view, expressing their conviction that Gemini 4 has successfully bridged the gap and caught up with the most advanced models in Silicon Valley.
On the financial front, Wall Street reacted cautiously to these leaks, with Alphabet shares dropping on Wednesday to trim their morning gains. After surging more than 2% before the Bloomberg report was published, gains receded to close up only about 0.5% right before the closing bell. This rapid decline reflected investors’ extreme sensitivity to any signals that might suggest Google is stumbling or losing momentum in the fierce AI race reshaping the balance of power among tech giants.
Frequently asked questions
Question: What is the reason behind the skepticism expressed by some Google employees toward the Gemini 4 model?
Answer: Employees reported a gap between the high scores achieved by the model in standard tests and the difficulty of its performance in complex real-world programming tasks.
Question: How did Google management and the DeepMind lab respond to those allegations?
Answer: Google denied this and affirmed the existence of an internal consensus on the model’s leadership and superior competitive capability, pointing to DeepMind leadership’s satisfaction with the results.
Question: How did financial markets and Alphabet shares react to these press leaks?
Answer: Alphabet stock trimmed its daily gains from over 2 percent to only about 0.5 percent at the close immediately after the report was published.
