Skip to main content

Google's Gemini 3.5 Pro Finally Gets a Date: July 17 Showdown with DeepSeek V4

At this year's I/O conference in May, Google made a big splash by announcing that its flagship large model, Gemini 3.5 Pro, would launch in June. But June came and went, and the model was nowhere to be found. During the delay, several core AI researchers jumped ship to competitors, and team morale hit rock bottom. Now, the wait is finally over: Gemini 3.5 Pro is set to officially drop on July 17.

Interestingly, mid-July is also when the domestic powerhouse DeepSeek V4 is expected to roll out its official version. The clash between these two titans—one from the West, one from the East—is shaping up to be one of the most electrifying moments in AI this year.

Rejecting 'Toothpaste Squeezing': A New Foundation, Trained from Scratch

So why the delay? Insiders say the Google team wasn't slacking off. Instead, they used the extra time to completely retrain the model from scratch, rather than just fine-tuning the previous Gemini 2.5 Pro. That means Gemini 3.5 Pro is a genuine leap forward, not a minor tweak. Google chose to risk missing deadlines to ensure the foundation was rock-solid—a sign of just how high their expectations are.

According to leaks, the biggest breakthrough lies in front-end generation. The model's design aesthetics have improved dramatically, with cleaner interfaces, stronger SVG graphics generation, and more concise initial front-end output. Game development also sees significant gains. But let's be real: Gemini 3.5 Pro probably won't dethrone Anthropic's Fable5 in coding capabilities—that crown is still firmly in place.

The Image Generation Star Returns: Nano Banana Pro Takes Aim at GPT-Image 2

Google's traditional strength with the Gemini series has always been its vast world knowledge—a point even DeepSeek has acknowledged, calling Google the strongest in that regard. Higher data quality means better overall performance, and with a new foundation paired with superior data, Gemini 3.5 Pro is expected to shine. But due to parameter scale limitations, it's unlikely to become the number one model overnight.

Still, Google has another ace up its sleeve: image generation. Its self-developed Nano Banana model once dominated the global leaderboard for ages, until OpenAI's GPT-Image 2 recently snatched the top spot. Now, Google is launching the next-generation Nano Banana Pro, built on Gemini 3.5 Pro, with a single goal: reclaim the throne in image generation and take on GPT-Image 2 head-to-head. A dual showdown—covering both text reasoning and image generation—is set to ignite in mid-July.

Image

Key Points

  • Launch Date: Gemini 3.5 Pro officially releases on July 17.
  • New Foundation: Trained from scratch, not fine-tuned, marking a significant upgrade.
  • Competition: Directly confronts DeepSeek V4, expected around the same time.
  • Image Generation: Nano Banana Pro aims to surpass GPT-Image 2.
  • Coding: Still trails behind Anthropic's Fable5.