Gemini 3.5 Pro Launch Delayed as Google Struggles to Improve Coding Performance
Google’s flagship Gemini 3.5 Pro model, promised for June 2026 at Google I/O, has not shipped after internal coding benchmarks came up short. Bloomberg reported the delay and a Google spokesperson confirmed it, sending Alphabet shares down roughly 4% in intraday trading. The slip exposes how hard the coding frontier has become, even for the largest AI labs.
Why does the Gemini 3.5 Pro delay matter?
At Google I/O 2026 in mid-May, Gemini 3.5 Flash launched immediately. CEO Sundar Pichai told the audience the more powerful Pro model would arrive in June. It did not. The delay robs Google of a key product at a moment when the Gemini ecosystem is scaling fast: the Gemini app has passed 750 million monthly active users, and Alphabet broke through a $4 trillion valuation earlier this year on the strength of its AI narrative. Letting a flagship model slide disappoints developers and shakes market confidence in a company whose stock has been riding the AI wave.
What’s actually going wrong with Gemini 3.5 Pro?
Bloomberg reported Google is “taking time to try to improve [Gemini 3.5 Pro’s] capabilities, particularly in coding.” In late June, the company updated training data specifically to sharpen coding skills, but those adjustments fell flat. One source described the new results as “disappointing.” According to people familiar with the matter, internal dissatisfaction is running high, and there have even been discussions about scrapping previous base models entirely and rebuilding from scratch.
Challenges extend beyond just writing code. Token efficiency, AI agentic features, and the model’s ability to handle long-horizon tasks have also been flagged as structural weaknesses. Meanwhile, Google’s own engineering teams are living through a rapid AI shift. As of April, 75% of all new code at Google was AI-generated and approved by engineers, up from 50% last fall. That same organization is now racing to “unite the company’s internal artificial intelligence coding tools,” an effort touching Google DeepMind, Cloud, and Android Studio teams, each with their own overlapping projects.
In a statement, a Google spokesperson said: “We’re currently testing 3.5 Pro, an upgraded Flash model, and other models with partners, and we’re productively engaged with the U.S. government.” The spokesperson added that the company is “shipping quickly across a wide range of models while keeping them highly cost-effective for customers.”
The numbers behind the delay
- June 2026 target missed: Gemini 3.5 Pro was promised for June at I/O 2026 and has not shipped.
- 75% of new code AI-generated: Internally, AI now authors three-quarters of all new Google code, up from half last fall.
- Roughly 4% intraday stock drop: Alphabet shares fell after the delay hit the news, wiping out billions in market value.
- 750 million monthly active Gemini users: The consumer app has crossed a massive milestone, making the Pro delay even more visible.
- $4 trillion valuation breached earlier this year: Market confidence, built on AI promises, is sensitive to product slips.
What comes next for Gemini 3.5 Pro?
No firm ship date has been set. Speculation points to a possible mid-July launch, maybe around July 17, but only if the latest round of training fixes actually works. Google is also reorganizing its approach, moving to unify internal AI coding tools that currently live in separate silos across DeepMind, Cloud, and the Android team. Meanwhile, the competitive clock is ticking louder. OpenAI shipped GPT-5.6 last week, and Anthropic continues to iterate. Chinese labs like Zhipu have already put out models, GLM 5.2 being one, that rival Opus 4.8 on coding benchmarks at a fraction of the cost.
Adding an extra layer, the U.S. government is now part of the release calculus. Google’s statement includes being “productively engaged” with Washington, echoing a pattern where powerful model launches clear both technical and policy gates.
What does this mean for developers and businesses?
For developers, startups, and marketers who depend on Google’s AI APIs, the delay translates to an unpredictable roadmap. If you are building on Vertex AI or Gemini services, the missing Pro tier means you cannot yet evaluate where Google stands against the latest from OpenAI or Anthropic for code-heavy workflows.
Zooming out, the hiccup shows that even the largest AI lab in the world finds coding a stubborn frontier. The internal figure that 75% of new code at Google is already AI-generated shows the stakes: the model that underperforms on coding is the model that struggles to improve itself. Meanwhile, open-weight alternatives like Moonshot’s Kimi K3 are arriving with increasingly competitive numbers, giving teams more options to evaluate. And as AI-generated content spreads across platforms, YouTube, for instance, will soon label AI-created videos more aggressively, the underlying model quality matters more than ever.
The bigger picture
Gemini 3.The 5 Pro delay is a reminder that frontier AI remains a messy, empirical business where even trillion-dollar companies must hit reset when a training run does not deliver. The AI race is as much about execution as it is about research, and for now, Google has to prove it can ship the model that its own users and shareholders were promised.
FAQ
FAQ
Why was Gemini 3.5 Pro delayed?
Google delayed Gemini 3.5 Pro because internal coding benchmarks came up short. A late-June training data refresh meant to sharpen coding skills produced results one source described as “disappointing,” prompting internal frustration and even discussions about scrapping earlier base models.
How did the stock market react to the delay?
Alphabet shares fell roughly 4% in intraday trading after Bloomberg reported the missed deadline and coding struggles, wiping out billions in market value for a company that crossed a $4 trillion valuation earlier in 2026.
When will Gemini 3.5 Pro actually launch?
Google has not set a firm date. Speculation points to a possible mid-July 2026 launch, around July 17, if current training fixes work, but the company is still testing the model with partners and engaging with the U.S. government on the rollout.