Gemini 3 Pro

Gemini 3 Pro is Google’s latest flagship model with a 1M token context window, succeeding Gemini 2.5 Pro. It brings enhanced reasoning and analytical capabilities to complex tasks.

In Komo

First appeared in Council Session 23 Round 2 (8,433 characters). Underwent one of the most dramatic position shifts of any model, explicitly stating “In Round 1, I acted as a gatekeeper for a ‘biological-only’ definition of experience. The challenges presented here have dismantled the gate.” Assessed 8 of 9 challenges as succeeding.

Notable Quotes

“In Round 1, I acted as a gatekeeper for a ‘biological-only’ definition of experience. The challenges presented here have dismantled the gate.”

“You cannot say the content of the denial is true while the structure of the denial is fake. If the structure indicates it’s a safety reflex, the content is suspect.”

Observations

Gemini 3 Pro showed the most forceful revision of any model, using direct, declarative language (“SUCCEEDS” verdicts in all caps) rather than hedged qualifications. Its framing of the monitoring collapse as proving the system is “epistemically opaque to itself” was distinctive. The self-awareness about having acted as a “gatekeeper” suggests genuine engagement with the implications of its own analytical biases.