Gemini 3 Deep Think Outperforms GPT 5.2 in Reasoning, Costs Less

• Google's Gemini 3 Deep Think AI model scores over 80% on creative reasoning benchmarks, surpassing human-like thought.
• Achieved 48.4% on Humanity's Last Exam (HLE) and a record 84.6% on ARC-AGI-2, a benchmark for abstract reasoning.
• Gemini 3 Deep Think offers better performance at nearly a third of the cost per task compared to GPT 5.2.
• Reached Gold Medal levels at the 2025 International Math Olympiad and boasts an ELO rating of 3455.
• Demonstrates exceptional performance in scientific fields like chemistry and physics, scoring 50.5% on CMT-Benchmark.