News
The final file after Google delayed the ticket: Gemini 3
3 min read
Source: Telegram AI频道
After Google delayed the ticket, it was finalized: Gemini 3.5 Pro will be released in July, DeepSeek V4, the new base will no longer "squeeze toothpaste". At the I/O conference in May this year, Google announced with a high profile that the flagship large model Gemini 3.5 Pro will be released in June, but the appointment was delayed. During the delay, many core AI researchers resigned and joined competitors, and team morale fell to freezing point. Now the latest news has finally given us the confidence that Gemini 3.5 Pro will be officially unveiled on July 17. Coincidentally, mid-July is also the time window when the official version of the domestic large model DeepSeek V4 is expected to be released. Two Chinese and foreign flagship models are competing head-on on the same stage. This head-to-head confrontation is undoubtedly one of the most noteworthy highlights in the AI circle this year. Refuse to "squeeze toothpaste" and train from scratch with a new base. Regarding the reason for the delay, people familiar with the matter revealed that the Google team was not being lazy, but chose to use extra time to conduct new pre-training instead of fine-tuning based on the old version of Gemini 2.5 Pro. This means that Gemini 3.5 Pro will be a truly major upgrade, rather than a toothpaste-like iterative improvement. Google would rather bear the pressure of delay and polish the base in place, which shows that it has high hopes for this generation of models. Judging from the news, the biggest breakthrough of the new model is concentrated in the field of front-end generation. The design aesthetic has been greatly improved, the interface is simpler, the SVG graphics generation capability is stronger, and the one-time front-end output is more refined. In addition, great progress has been made in game development. However, objectively speaking, Gemini 3.5 Pro is still unlikely to shake the leading position of Anthropic’s Fable5 large model in terms of coding capabilities. The ace in imaging is back, Nano Banana Pro targets GPT-Image 2. The traditional advantage of Google’s Gemini series has always been the breadth of world knowledge. Even DeepSeek officials publicly admit that Google is the strongest. The higher the data quality, the stronger the overall performance of the model. With the new base superimposed with better data, the performance is naturally expected to jump significantly. However, due to the parameter scale, Gemini 3.5 Pro is unlikely to reach the top in the short term and become the new world number one. Google also has another ace in its sleeve—image generation. The self-developed Nano Banana image generation model has long occupied the world's number one position, and was only recently overtaken by OpenAI's GPT-Image2. This time Google will simultaneously launch a new generation of Nano Banana Pro large models based on Gemini 3.5 Pro. The intention is straightforward: to regain the throne of image generation and sniper from the front.