Google Delays Gemini 3.5 Pro Launch as Model Falls Short on Coding Benchmarks
Alphabet shares fell sharply after a Bloomberg report revealed Gemini 3.5 Pro is months behind schedule, with Google racing to fix coding shortfalls before a wider rollout.
The most powerful AI model by Google isn’t just ready yet. Alphabet’s Google is months behind schedule on releasing Gemini 3.5 Pro, its flagship large language model, as the company works to fix shortcomings in coding performance, Bloomberg reports.
Alphabet shares fell roughly 4% Thursday on the news, erasing close to $200 billion in market value in a single session. The delay has reportedly frustrated engineers and researchers internally as rivals OpenAI and Anthropic close the gap.
Coding Shortfalls Behind the Delay
Reuters reports the model was originally due in June, according to comments Pichai made during Google’s I/O developer conference in May.
Bloomberg reports that Google updated the data used to train Gemini late last month specifically to sharpen its coding skills, but the results fell short of internal expectations.
The timing is awkward: both OpenAI and Meta recently shipped new models, namely GPT-5.6 and Muse Spark 1.1, that further outpace Google’s current lineup at generating software code, an area that has become one of the clearest battlegrounds among AI labs.
A Google spokesperson responded to the report by saying the company is shipping quickly across a wide range of models, according to Yahoo Finance, adding that Gemini 3.5 Pro, an upgraded Flash model, and other systems are currently being tested with partners.
He added that Google is also working with the US government on AI model testing and broader safety frameworks.
Internal Friction Slowing Google’s AI Roadmap
The delay reflects deeper organizational strain, according to Yahoo’s reporting on Bloomberg, which cited 10 current and former Google employees.
Preparing a release model involves multiple stakeholders as Google integrates AI across Search, Maps, and YouTube, a process sources said routinely slows the launches.
Google Cloud, Google DeepMind, and the Android team have also developed separate AI coding tools, sometimes with overlapping consumer products, while some engineers oppose AI-written code, arguing it should remain human-authored to meet Google’s standards.
Earlier restrictions preventing engineers from using Gemini to write or analyze proprietary code over data leak concerns have since been relaxed.
Google says 75% of its code is now AI-generated and reviewed before production. It has also consolidated many coding tools under Google Antigravity, the team behind the Gemini Mac app launch, with Chief AI Architect Koray Kavukcuoglu leading efforts to unify the company’s AI coding systems.
Market Reaction and What’s Next
CNBC reports Gemini 3.5 Pro was first unveiled in May at Google I/O and has already been used internally, though its public commercial launch kept slipping.
Customer reaction to Gemini 3.5 Flash, launched at the same event, has been mixed. Yahoo Finance reports Figma product manager Rodrigo Davies praised its balance of speed and quality in Figma’s new AI agent.
However, Platzi CEO Freddy Vega said it is slower, more expensive, and less capable than Google’s 3.1 Flash model and premium rivals, leading his team to Anthropic’s Claude 3.5 Sonnet for reasoning-intensive tasks.
Alphabet reports second-quarter earnings Wednesday, July 22, when investors will look for updates on Cloud growth and Google’s AI spending plans heading into 2027.
Source: Google Gemini Launch Delayed as Tech Falls Short of Internal Goals
