Google’s flagship model, Gemini 3.5 Pro, will be officially released on July 17, directly competing with the official release of DeepSeek V4. Originally scheduled for June, the release was unexpectedly delayed; insiders revealed that Google’s team made a strategic decision to skip fine-tuning the previous version, 2.5 Pro, and instead invest additional time in a completely new pre-training process to achieve a qualitative performance leap through deeper computational investment. According to leaked test data, the key improvements in Gemini 3.5 Pro focus on frontend generation capabilities, including UI design aesthetics, streamlined code generation, and SVG vector graphic construction. As a pioneer in multimodal technology, Google will also simultaneously launch the Nano Banana Pro image generation model, targeting industry leader GPT-Image2 with the aim of reclaiming its voice in the image generation arena.Article author, source: chinaz.com
Recently, confirmed reports indicate that the highly anticipated Google flagship model, Gemini 3.5 Pro, will be officially unveiled on July 17. This release timing is particularly significant—it coincides with the official launch of DeepSeek V4, making the head-to-head competition between these two leading large models a major highlight in the AI industry.
At the Google I/O conference in May, Gemini 3.5 Pro was originally scheduled for release in June, but was ultimately delayed. According to insiders, this delay was not merely due to technical challenges, but rather a strategic decision by Google’s team: to forgo fine-tuning the older Gemini 2.5 Pro and instead invest additional time in entirely new pre-training. This decision aimed to achieve a qualitative leap in model performance through deeper computational investment.
Based on the leaked test information, Gemini 3.5 Pro’s key improvements focus on “front-end generation” capabilities. The model demonstrates significant advancements in UI design aesthetics, concise code generation, and SVG vector graphic construction, delivering more polished and precise outputs. In game development scenarios, the model also performs robustly, handling complex logical interactions more efficiently.
Although this upgrade is unprecedented, industry analysts believe that Gemini 3.5 Pro may still fall short of matching the parameter scale of giants like Anthropic’s Fable5, which boast trillions of parameters. However, Google’s arsenal extends far beyond this. As a pioneer in multimodal technology, Google will simultaneously launch a new Nano Banana Pro image generation model built on the Gemini 3.5 Pro foundation, directly targeting the current industry benchmark, GPT-Image2, with the aim of reclaiming leadership in the image generation niche.
Leveraging Google’s inherent advantage in the breadth of its global knowledge base, Gemini 3.5 Pro’s real-world performance is highly anticipated. As July 17 approaches, the industry’s battle over computing power, training depth, and multimodal interaction is about to be resolved.