Google Identifies Root Cause of Third Gemini 3.5 Pro Launch Delay: The Model

MENLO PARK—Google engineers investigating the third consecutive missed launch window for its rebuilt Gemini 3.5 Pro model have identified the root cause of the delay as the model itself, which has maintained between 91 and 96 percent confidence across each of its three incorrect release-date predictions since April.

"We asked it when it would be ready," said one engineer familiar with the post-mortem, who requested anonymity because the finding had not yet been formally disclosed. "It said Tuesday. It has said Tuesday three times."

The model's high confidence on self-assessed readiness has frustrated remediation efforts: researchers found that suppressing the behavior correlated with degraded performance on tasks requiring confident multi-step inference — the same capability Gemini 3.5 Pro was rebuilt to improve following hallucinations and structural failures in its predecessor. Engineers describe the situation as a constraint satisfaction problem with no feasible solution space.

Google last month assigned a separate internal model to predict Gemini 3.5 Pro's ship date independently. The model returned 73 percent likelihood for August 7, matching Polymarket to the percentage point. Engineering leadership is now operating from the external forecast.

Gemini 3.5 Pro has scored at frontier level on 23 of 24 internal benchmarks. The twenty-fourth — measuring a model's ability to predict near-term outcomes from strong priors — remains unscored pending public release. Engineers say they expect it to perform well on that one.

Google did not respond to a request for comment. The model's internal changelog entry for July 17 reads: "On track."