Gemini 3.5 Pro Rolls Out After Missing Two GA Deadlines

Gemini 3.5 Pro started rolling out in July, missing both its May and June GA targets. Reporting from the-agent-report.com traces the slip to three problems flagged by early testers. Token efficiency lagged, coding performance fell short of Google's own flagship bar, and long-horizon reasoning didn't match the I/O demo.

The timing is rough for Google. The delay landed in the same week as four senior DeepMind exits. Noam Shazeer left for OpenAI. Nobel laureate John Jumper, plus Jonas Adler and Alexander Pritzel, left for Anthropic. Alphabet's market cap dropped roughly $225B over the same stretch, per the report.

The delay matters less than its cause. Token efficiency, coding quality, and long-horizon reasoning are what breaks first when a model runs multi-step agent workloads instead of single prompts. A flagship model missing its own GA date on exactly those three benchmarks says something. Agentic performance is the real test for frontier models now, ahead of one-shot answer quality.

Builders evaluating Gemini 3.5 Pro for agent pipelines should test token consumption and reasoning stability across long tool-call chains before trusting the launch marketing. The report doesn't say which of the three problems is fixed in this rollout.