Google (GOOGL.US) launches a new version of Gemini Flash, but the flagship 3.5 Pro “difficult to produce” is increasing concerns about cutting-edge competitiveness

The Zhitong Finance App learned that Google (GOOGL.US) has launched another version of its main Gemini Flash artificial intelligence (AI) model, but has not revealed when the Gemini 3.5 Pro, a more powerful AI model currently plagued by delays, will be released.

Google said in a blog post on Thursday that the newly launched Gemini 3.7 Flash performs better than previous generation products in programming tasks such as debugging, and is more capable of producing code that can be deployed and used in production environments when first generated. The company added that the new model can also complete application development with fewer prompts, provide a better developer experience, and reduce the price of tokens required to run the model. Gemini Spark, Google's AI productivity intelligence agent, will be powered by 3.7 Flash starting Thursday.

Google said that in order to fulfill its “Frontier Safety” (Frontier Safety) promise, the company has also added more powerful security features to Gemini 3.7 Flash, including blocking malicious hacker attacks and preventing dangerous chemical, biological, radiological, or nuclear materials from being misused without interfering with safe and beneficial normal operation.

Despite the continuous expansion of the Gemini Flash product line, it is still difficult for Google to keep up with OpenAI and Anthropic in the high-risk competition to build the most advanced AI model. The extension of Gemini 3.5 Pro has already caused investors to question Google's product roadmap, particularly in areas with important commercial value such as AI programming.

At the 2026 I/O developer conference held in May of this year, Pichay said that the Gemini 3.5 Pro model would be launched in June, but the model has not been launched yet. According to the original plan, the Gemini 3.5 Pro model should take on the task of re-impacting Google's cutting-edge AI competition. After all, in the context of OpenAI and Anthropic's continuous updating of model capabilities, the Gemini Pro series has always been an important yardstick for the outside world to measure Google's AI strength. However, the delay in deciding when the next flagship model will be released has further fueled widespread doubts about whether the tech giant can surpass its rivals and successfully turn huge AI investments into market-leading tools and services.

Google CEO Sundar Pichai said during the company's last earnings call in July that Google plans to launch the model at a faster rate, and that the company has invested significant computing resources to train the upcoming Gemini 4 model. Just before the earnings release, the tech giant also launched three Flash versions with the goal of achieving greater efficiency and quality.

Outsiders speculate that the actual capabilities of the Gemini 3.5 Pro model may not have met Google's initial expectations. Industry analysis agency Semi Analysis believes that the capabilities of the Gemini 3.5 Pro are about close to the level of Anthropic Claude Opus 4.5, and the latter is already a model released at the end of November last year. The agency even said in the latest report that Google may have shelved the Gemini 3.5 Pro model internally.

According to people familiar with the matter, Google has been spending time trying to improve the capabilities of the Gemini 3.5 Pro, especially in terms of programming, which has caused the launch of this model to be delayed by several months. According to 10 current and former employees, the delay has already frustrated Google engineers, AI researchers, and management. Many of them worry that as Anthropic and OpenAI continue to launch models that surpass Gemini's capabilities, Google may lose its edge in market competition. People familiar with the matter said that Google involved multiple levels of stakeholders in the preparation process for model release, and also made efforts to integrate AI into a huge product system, including search, maps, and YouTube, which may cause delays in the release process.

Top talents are leaving one after another, and the AI leadership has been reshuffled! Google has been “not so happy” lately

In addition to the extension of the Gemini 3.5 Pro model, Google's recent brain drain is also worrying. Earlier this month, Chief Scientist Jeff Dean announced his departure after 27 years at the helm. Previously, a number of well-known researchers have left Google, including Norm Chazelle, one of the authors of the 2017 landmark paper “Attention Is All You Need” — an article that laid the foundation for generative AI, and now all eight authors have left Google. Chazel switched to OpenAI in June of this year, and it is less than two years since Google recalled it through “takeover recruitment” for nearly $3 billion. Shortly after his departure, Nobel laureate John Jamper also left DeepMind to join Anthropic.

D.A. Davidson analyst Jill Luria pointed out that Google's top talent drain is clearly trending. He said, “They're not passionate about commercializing AI; they want to be part of history. So they see Anthropic, OpenAI, or other startups as places where history can be written.”

Google's investment in global data centers, chips, and related infrastructure is almost unrivaled, yet computing power is still in short supply. Each TPU is assigned to train models, support Google products, or fulfill cloud customer contracts, representing a choice between multiple priorities. According to a number of people familiar with the matter, who did not wish to be named, some Google researchers are increasingly dissatisfied with obtaining computing power — it is difficult for them to obtain the computing resources needed to advance cutting-edge projects, yet they have seen Google sell self-developed TPUs to external customers, including Anthropic. Furthermore, Google's internal hierarchical approval is cumbersome, and the transformation of research results into products requires multiple levels of clearance, making OpenAI, Anthropic, and even younger startups more attractive to AI developers — they prefer lab work over financial data.

However, Google's AI leadership has faced large-scale adjustments, which seem to indicate that the company is ready to meet the challenge. According to reports, DeepMind co-founder Demis Hassabis handed over day-to-day management rights, stepped down as the overall head of commercialization at Gemini, and transferred to the position of Chairman of DeepMind and Chief Scientist of Google, focusing mainly on long-term AI research. Former DeepMind Chief Technology Officer Corey Kavukkuoglu was promoted to Senior Vice President of DeepMind (DeepMind no longer has an independent CEO), responsible for the development and operation of the Gemini model, and reports directly to Pichay.

Furthermore, Google co-founder Sergei Brin will be more directly involved in Gemini. Currently, Brin has no formal executive positions. However, since the release of ChatGPT, he has re-intervened in Google's daily AI affairs, participated in model testing, discussed technical routes, and exerted more and more influence on Gemini's development direction.

The core of this restructuring is the relocation of the AI decision-making center from London back to Silicon Valley, aimed at reversing trends that have plagued the company since at least 2023. Back then, Google merged two highly regarded science labs that originally operated independently: Google Brain, which is based at the company's Mountain View headquarters, and DeepMind, which is rooted in London. Although the two laboratories merged under the name of Google DeepMind, the researchers are still working on different continents. According to people familiar with the matter, this arrangement made the decision-making process more complicated, and also made talents from both places feel dissatisfied.

A major focus of Google's future business strategy is to transform the “research federation” that used to be scattered in the traditional systems of DeepMind in London and Google Brain in California into a Gemini product delivery machine with Mountain View as the core and directly responsible to Pichay. Kavukkuolu is also in charge of Google DeepMind's day-to-day operations and chief AI architect responsibilities, which means model pre-training, post-training, evaluation, computing power scheduling, and collaboration with Cloud, Search, and developer products will be incorporated into a shorter decision chain.

These latest personnel adjustments and transfers do not weaken basic research, but rather split management of “long-term scientific exploration” and “quarterly product delivery” in an attempt to solve the problems of slow decision-making and slow commercialization of research results by cross-continental teams after the 2023 merger.

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

But now the trickier task is to quickly transform model capabilities into developer tools, enterprise services, and stable revenue. In the AI programming and enterprise markets, OpenAI and Anthropic have established a first-mover advantage. Google's senior management and board of directors are worried that the company's research strength has yet to be fully transformed into product competitiveness.

Gemini previously carried multiple functions such as research, modeling, application, and safety at the same time. The decision-making chain is long, and it is easy for research goals to clash with the pace of the product. The new management structure hands over Gemini's daily development to Kavukkuolu, helping to reduce the decision-making cycle and speed up model release and product implementation.

The report also mentioned that some Google executives are dissatisfied with the extent to which Hassabis invests in commercialization work. AlphaFold is considered one of the controversial cases. This protein structure prediction system helped Hassabis win the 2024 Nobel Prize in Chemistry, and also had a huge impact on scientific research. However, after the free opening of the project, Google did not receive commercial returns commensurate with the scale of investment. People close to Google denied that there was an obvious conflict between the two parties. They note that Hassabis led the early launch of Gemini and has been concerned about the progress of the project.

After major changes in Google's AI leadership, whether the new structure works depends on the smooth coordination of the three routes. Research needs to preserve room for exploration, model teams need to iterate faster, and product departments need to find users who are really willing to pay. For Google, the list of top papers and models is no longer enough to dispel external doubts; the next competition will be on coding tools, corporate markets, product experience, and commercial revenue.

Brin's return to the center of AI power shows that Google already sees Gemini as a core battle requiring the founder's personal intervention. Hassabis's retreat from the front line of scientific research also means that the research-led model that DeepMind has formed over the past ten years is changing. Google has models, computing power, data, distribution channels, and a huge base of enterprise customers. Today, it needs to prove that it can compress these resources into a faster product link.