Google Accelerates on Economic AI but Delays Gemini 3.5 Pro

Google Unveils New AI Models While Delaying Gemini 3.5 Pro

Google recently announced the launch of several new artificial intelligence models in its Gemini lineup, while also revealing details about the development of Gemini 4. Among these new offerings, three models stand out for their speed and reduced operational costs for AI agents. However, the highly anticipated model, Gemini 3.5 Pro, initially scheduled for release in June, is still in the testing phase. Meanwhile, Google has embarked on a crucial stage in the creation of Gemini 4, which is set to be the company's next flagship model.

More Efficient and Cost-Effective Models

Among the new models, the 3.6 Flash stands out as a true "workhorse," surpassing the previous model, the 3.5 Flash, particularly in complex tasks such as programming. According to the Artificial Analysis Index, a recognized benchmark for evaluating AI tools, the 3.6 Flash manages to reduce token usage—these units of text processed by the AI—by 17% compared to its predecessor.

Google has also introduced the 3.5 Flash-Lite, touted as the fastest and most cost-effective version of the 3.5 series to date. Additionally, the 3.5 Flash Cyber has been designed to detect and correct cybersecurity vulnerabilities, offering performance comparable to its competitors but at a lower cost per token. Tulsee Doshi, Senior Director of Product Management for the Gemini group at Google, emphasized in a blog post that this model will first be available to governments and select trusted partners.

Competition and Cost-Reduction Strategy

Google's strategy comes as other major players like Anthropic and OpenAI have recently launched their own cybersecurity models. Google appears to be betting on cost reduction to differentiate itself from the competition. The Flash Cyber model, with its lower cost, could provide Google with a strategic advantage, especially in a context where companies are looking to limit their AI spending.

Google's CEO, Sundar Pichai, had already pointed out earlier this year that companies are quickly reaching their annual token budgets. He suggested that a mix of models like Flash could lead to significant savings. Google claims that its new models offer an optimal balance between efficiency and power for the operation of AI agents.

Delay of the Gemini 3.5 Pro Model

The Gemini 3.5 Pro model, which was supposed to be Google's next major model, is still in the testing phase with certain partners. Google has indicated that it will be launched "as soon as it is ready." In June, Business Insider reported that the release of the 3.5 Pro had been pushed to July, and Bloomberg recently suggested that the model could face further delays.

Currently, Google does not rank among the top ten on the Artificial Analysis leaderboard, which evaluates models based on criteria such as mathematics and reasoning. However, Google announced that it has launched its "most ambitious pre-training run to date" for the development of Gemini 4. While the launch timeline for the 3.5 Pro remains uncertain, Google hopes that its efforts to reduce costs and optimize token usage will allow it to regain a leadership position in the AI market.