Gemini 3 Deep Think — benchmark results

Google Gemini 3 variant using Deep Think mode for harder multi-step reasoning tasks. Provider: Google. Released 2026-02-26. Access: API.

Unified ELO 2186 ± 71, rank #4 of 1776 rated models, from 20 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
Google Gemini 3 Deep Think - ARC-AGI-284.6Score (%)100
Google Gemini 3 Deep Think - CMT-Benchmark50.5Pass@8 (%)100
Google Gemini 3 Deep Think - Codeforces3455Elo100
Google Gemini 3 Deep Think - GPQA Diamond93.8Score (%)100
Google Gemini 3 Deep Think - Humanity's Last Exam (no tools)48.4Score (%)100
Google Gemini 3 Deep Think - Humanity's Last Exam (search and code)53.4Score (%)100
Google Gemini 3 Deep Think - International Chemistry Olympiad 2025 (theory)82.8Score (%)100
Google Gemini 3 Deep Think - International Math Olympiad 202581.5Score (%)100
Google Gemini 3 Deep Think - International Physics Olympiad 2025 (theory)87.7Score (%)100
Google Gemini 3 Deep Think - MMMU-Pro81.5Score (%)100
LiveCodeBench Pro3298Rating (CF-style)100
CritPt25.7Accuracy (self-reported)98.5

Interactive version: theaggregate.ai/model?slug=gemini-3-deep-think · How the rankings work · Data refreshed daily, snapshot 2026-07-22.