At the IO conference in May, Google announced that the flagship large model Gemini 3.5 Pro would be released in June, but it was eventually delayed. During this period, many AI experts resigned and joined rival companies, and morale was greatly reduced. The latest news says,The Gemini 3.5 Pro large model will be released on July 17.What’s interesting is that mid-July happens to be the day when the official version of DeepSeek V4 is released. The competition between two flagship large models at home and abroad is worthy of attention.
The reason why Gemini 3.5 Pro was delayed was that the team wanted to use the extra time to conduct new pre-training instead of fine-tuning based on the old Gemini 2.5 Pro large model.
According to this revelation, the improvements of Gemini 3.5 Pro are mainly in front-end generation, which has made a major leap forward, with better design taste, simpler UI, more powerful SVG generation, and more refined one-time front-end results.
There has also been great progress in game development.But it’s still impossible for the Gemini 3.5 Pro to surpass Anthropic’s Fable 5 models.
The Gemini 3.5 Pro large model will be a major upgrade this time. Google will not make improvements based on the previous Gemini 2.5 Pro. It should be a brand new large base model. It would rather delay the ticket than spend more time training, so there will be a significant improvement in performance.
Google’s Gemini advantage has always been the breadth of world knowledge. Even DeepSeek officials admit that Google is the strongest in this regard. In theory, the better the data quality, the performance of Gemini 3.5 Pro will be greatly improved.
However, it is unlikely that Gemini 3.5 Pro will become the new world number one all of a sudden. A's Mythos and Fable large models are said to have a scale of 5 trillion or even 10 trillion parameters, which are extremely expensive. Gemini 3.5 Pro will not be able to achieve such a large scale, and there is naturally a gap in the upper limit.
But Google also has an advantage, and that is image generation. The Nano Banana graph model has been the world's number one for a long time, and was only surpassed by OpenAI's GPT-Image 2 not long ago.This time Google will launch a new Nnao Banana Pro large model based on Gemini 3.5 Pro, which is expected to compete more directly with GPT-Image 2.
