Google's flagship model, Gemini 3.5 Pro, has been delayed for months. Originally slated for a June launch, the model's coding capabilities have consistently fallen short of internal targets. Even after updating training data in late June, results remain unsatisfactory.
According to APPSO, which cites insiders, the latest evaluations of Gemini 3.5 Pro might actually be worse than Meta's Muse Spark. Access to testing has recently been revoked for employees in related business units, leaving only the model team able to continue using and reviewing evaluation results.
APPSO also reports that morale in Google's Gemini teams is low. Some employees, concerned about team restructuring and layoffs, are considering joining other AI labs. Recent departures include Noam Shazeer moving to OpenAI and AlphaFold core researcher John Jumper leaving for Anthropic.
That said, there is no solid evidence that Gemini 3.5 Pro has been canceled. Google responded that the model is still being tested with partners, and that they are also testing an upgraded version of Flash and other models.