Google launches the new version of Gemini Flash, but the flagship model 3.5 Pro is "difficult to produce," exacerbating concerns about its competitive capabilities at the forefront

According to Zhitong Finance APP, Google (GOOGL.US) has launched another version of its flagship Gemini Flash artificial intelligence (AI) model, but has not disclosed when the more powerful AI model Gemini 3.5 Pro, which is currently facing delays, will be released.

In a blog post on Thursday, Google stated that the newly launched Gemini 3.7 Flash outperforms its predecessor in programming tasks such as debugging, and has a stronger ability to generate deployable, production-ready code on the first attempt. The company added that the new model can complete application development with fewer prompts, providing a better developer experience and reducing the cost of tokens required to run the model. Google's AI productivity agent Gemini Spark will be supported by 3.7 Flash starting Thursday.

Google stated that to fulfill its "Frontier Safety" commitment, it has added more robust security features to Gemini 3.7 Flash, including preventing malicious hacking attempts and stopping the misuse of dangerous chemical, biological, radioactive, or nuclear materials, while not hindering safe and beneficial normal operations.

Despite continuously expanding the Gemini Flash product line, Google is still struggling to keep pace with OpenAI and Anthropic in the high-stakes competition to create cutting-edge AI models. The delay of Gemini 3.5 Pro has led investors to begin questioning Google's product roadmap, especially in areas with significant commercial value such as AI programming.

At the I/O developer conference held in May this year, Sundar Pichai stated that the Gemini 3.5 Pro model would be launched in June, but the model has yet to go live. According to the original plan, the Gemini 3.5 Pro model was supposed to take on the task of helping Google re-enter the forefront of AI competition. After all, in the context of OpenAI and Anthropic continuously enhancing model capabilities, the Gemini Pro series has been an important benchmark for measuring Google's AI strength. However, the uncertainty surrounding the release date of the next flagship model has further intensified doubts about whether this tech giant can surpass its competitors and successfully convert its massive AI investments into market-leading tools and services.

Google CEO Sundar Pichai stated during the company's last earnings call in July that Google plans to launch models at a faster pace and that the company has invested significant computational resources to train the upcoming Gemini 4 model. Just before the earnings report was released, the tech giant also launched three Flash versions aimed at achieving higher efficiency and quality.

There is speculation that the actual capabilities of the Gemini 3.5 Pro model may not have met Google's initial expectations. Industry analysis firm Semi Analysis believes that the capabilities of Gemini 3.5 Pro are roughly comparable to those of Anthropic Claude Opus 4.5, which is a model released at the end of November last year. The firm even stated in its latest report that Google may have shelved the Gemini 3.5 Pro model internally According to informed sources, Google has been spending time working to enhance the capabilities of Gemini 3.5 Pro, particularly in programming, which has led to the model's launch being delayed by several months. Ten current and former employees revealed that this delay has frustrated Google engineers, AI researchers, and management, many of whom are concerned that as Anthropic and OpenAI continue to release models that surpass Gemini's capabilities, Google may lose its competitive edge in the market. Sources indicate that Google involves multiple layers of stakeholders in the model release preparation process while also striving to integrate AI into its vast product ecosystem, including Search, Maps, and YouTube, which may contribute to the delay in the release process.

Top Talent Exits and Major Reshuffling in AI Leadership! Google Recently "Unsettled"

In addition to the delay of the Gemini 3.5 Pro model, the recent talent exodus at Google is also concerning. Earlier this month, Chief Scientist Jeff Dean announced his departure after 27 years of service. Several prominent researchers have already left Google, including Noam Shazeer, one of the authors of the landmark 2017 paper "Attention Is All You Need," which laid the foundation for generative AI; all eight authors of that paper have now left Google. Shazeer joined OpenAI in June of this year, less than two years after Google brought him back through "acquisition-style hiring" for nearly $3 billion. Shortly after his departure, Nobel laureate John Jumper also left DeepMind to join Anthropic.

D.A. Davidson analyst Gil Luria pointed out that the trend of top talent leaving Google is evident. He stated, "They are not keen on commercializing AI; rather, they want to be part of history. Therefore, they view Anthropic, OpenAI, or other startups as places where they can write history."

Google's investment scale in global data centers, chips, and related infrastructure is nearly unmatched, yet there is still a shortage of computing power. Each TPU allocated for training models, supporting Google products, or fulfilling cloud customer contracts represents a choice among multiple priorities. According to several unnamed insiders, some Google researchers are increasingly dissatisfied with the acquisition of computing power—they find it difficult to obtain the computational resources needed to advance cutting-edge projects while seeing Google sell its self-developed TPUs to external customers, including Anthropic. Additionally, Google's internal approval processes are cumbersome, and transforming research results into products requires multiple layers of approval, making OpenAI, Anthropic, and even younger startups more attractive to AI researchers—they prefer laboratory work over financial report numbers.

However, Google's AI leadership has undergone a major restructuring, seemingly indicating that the company is preparing to meet challenges. It is reported that DeepMind co-founder Demis Hassabis has handed over daily management responsibilities and stepped down as the overall head of Gemini commercialization, taking on the role of DeepMind chairman and Google Chief Scientist, focusing primarily on long-term AI research. Former DeepMind Chief Technology Officer Koray Kavukcuoglu has been promoted to Senior Vice President of DeepMind (DeepMind will no longer have an independent CEO), responsible for the development and operation of the Gemini model And report directly to Pichai.

In addition, Google co-founder Sergey Brin will be more directly involved in Gemini. Currently, Brin does not hold an official executive position. However, since the release of ChatGPT, he has re-engaged in Google's daily AI affairs, participating in model testing, discussing technical directions, and exerting increasing influence on the development direction of Gemini.

The core of this reorganization is to move the AI decision-making center back from London to Silicon Valley, aiming to reverse a trend that has plagued the company since at least 2023. That year, Google merged two originally independent and highly valued scientific laboratories: one was Google Brain, located at the company's Mountain View headquarters, and the other was DeepMind, rooted in London. Although the two labs merged under the name Google DeepMind, researchers still worked on different continents. According to insiders, this arrangement complicated the decision-making process and left talent in both locations dissatisfied.

A major focus of Google's future business strategy is to transform the "research federation" that was previously scattered between London’s DeepMind and California’s Google Brain traditional system into a Gemini product delivery machine centered in Mountain View and directly accountable to Pichai. The dual role of Kavukcuoglu overseeing both the daily operations of Google DeepMind and serving as Chief AI Architect means that model pre-training, post-training, evaluation, computing power scheduling, and collaboration with Cloud, Search, and developer products will be included in a shorter decision-making chain.

These latest personnel adjustments and movements do not weaken basic research but rather attempt to manage "long-term scientific exploration" and "quarterly product delivery" separately, trying to address the issues of slow decision-making across continents and slow commercialization of research results that arose after the merger in 2023.

The Gemini 3 series models and Nano Banana image editing tool have helped Google regain market attention. Google has also proven that it still possesses the capability to develop cutting-edge models.

However, the more challenging task at hand is to quickly translate model capabilities into developer tools, enterprise services, and stable revenue. In the AI programming and enterprise market, OpenAI and Anthropic have already established a first-mover advantage. Google executives and the board are concerned that the company's research strength has not yet fully translated into product competitiveness.

Previously, Gemini simultaneously carried out multiple functions including research, modeling, applications, and security. The long decision-making chain made it easy for research goals and product rhythms to conflict. The new management structure assigns the daily development of Gemini to Kavukcuoglu, which helps shorten decision cycles and accelerate model releases and product implementations.

Reports also mention that some Google executives are dissatisfied with Demis Hassabis's level of involvement in commercialization efforts. AlphaFold is seen as one of the controversial cases. This protein structure prediction system helped Hassabis win the 2024 Nobel Prize in Chemistry and has had a significant impact on scientific research, but after the project was made freely available, Google did not receive commercial returns that matched the scale of its investment. People close to Google deny that there are significant conflicts between the two parties. They point out that Hassabis led the early release of Gemini and has always been concerned about the project's progress After significant changes in Google's AI leadership, whether the new structure can be effective depends on the smooth coordination of three routes. Research needs to retain exploratory space, the model team needs to improve iteration speed, and the product department must find users who are truly willing to pay. For Google, top papers and model rankings are no longer sufficient to dispel external doubts; the upcoming competition will focus on code tools, enterprise markets, product experience, and commercial revenue.

The return of Brin to the center of AI power indicates that Google views Gemini as a core battle that requires the personal involvement of the founders. Demis Hassabis stepping back to a research frontline also means that the research-led model developed by DeepMind over the past decade is undergoing changes. Google possesses models, computing power, data, distribution channels, and a large enterprise customer base. Now, it needs to prove that it can compress these resources into a faster product pipeline