The Gemini 3.5 Pro delay follows reports that Google’s model fell short of internal goals, particularly in coding. Google introduced Gemini 3.5 Flash at I/O 2026 and said the Pro version would arrive in June. That deadline passed without a public launch, leaving the company to continue testing and improving the model.
Coding performance is at the centre of the delay
According to reporting cited by 9to5Google, Google is taking more time to improve Gemini 3.5 Pro’s capabilities, with coding identified as a particular concern. The company reportedly updated training data in late June in an attempt to strengthen coding skills, but the results were disappointing. That suggests development encountered a more fundamental problem than a routine launch adjustment.
Google has not published a full technical explanation of the delay. It is also unclear how the latest model performs in other areas such as reasoning, writing or multimodal tasks. The public information points to coding as the immediate obstacle, not necessarily a broad failure across every benchmark.
Google says several models are still being tested
In a statement, Google said it is currently testing Gemini 3.5 Pro, an upgraded Flash model and other models with partners. The company also emphasized that it is shipping quickly across a wide range of models while trying to keep them cost-effective for customers.
That wording suggests Google may be considering more than one path forward. Gemini 3.5 Pro could launch after additional training, while an upgraded Flash model may offer a faster or cheaper alternative. For developers, the important question will be whether Google can make the model more capable without pushing up latency and usage costs.
AI coding is already part of Google’s internal workflow
The delay comes as Google expands its own use of AI-assisted programming. The company has said that 75 percent of new code at Google is now AI-generated and approved by engineers, up from 50 percent the previous autumn. That figure makes coding performance especially important for Google’s model strategy: the company is both building AI coding tools and relying on them internally.
Reports also describe capacity constraints affecting internal AI coding tools and an effort to unite separate coding systems across the company. Google DeepMind, Google Cloud, Android Studio and AI Studio have all been developing related tools, which creates opportunities but also makes consistency harder to maintain.
What the delay means for users
For ordinary Gemini users, the delay means there is no confirmed 3.5 Pro release date to plan around. For developers, it is a reminder that model labels and conference timelines do not guarantee production readiness. Coding assistants need to be evaluated on real repositories, debugging tasks and long-running projects, not only polished demonstrations.
Google says testing continues. Until the company publishes a launch date and performance details, Gemini 3.5 Pro should be treated as an unfinished model rather than a product that is simply waiting for a scheduled rollout.
Why coding benchmarks do not tell the whole story
A model can score well on short programming benchmarks and still struggle inside a large production repository. Real development work requires the model to follow project conventions, understand dependencies, preserve tests and avoid introducing subtle regressions. Long-context reasoning and reliable tool use are therefore as important as generating a correct function in isolation.
Google’s reported decision to revise training data indicates that the company is looking beyond a cosmetic fix. If the model’s coding behaviour is inconsistent, additional training and evaluation may be safer than releasing it on the original schedule and correcting problems after developers begin building around it.
What developers should watch before launch
When Gemini 3.5 Pro eventually arrives, developers should compare its accuracy, latency, context limits and pricing with the upgraded Flash model. They should also test it against private repositories and realistic debugging tasks before adopting it for sensitive workflows. Human review remains essential, especially where generated code affects security, payments or customer data.
Follow the iDigitalNews AI section and our Google coverage for launch details, model access and developer pricing when Google makes them public.
The missed June target also raises expectations for Google’s next announcement. A revised launch should include transparent comparisons with Gemini 3.1 Pro and the upgraded Flash model, especially on software-engineering tasks. Developers will want evidence of better repository-level reasoning, fewer incorrect edits and dependable performance over long sessions before treating the new Pro model as a meaningful upgrade.
Source: 9to5Google
