Gemini 2.5 Pro (Preview) — benchmark results

Google preview Gemini 2.5 Pro model, an early Pro snapshot from the Gemini 2.5 family. Provider: Google. Released 2025-03-25. Access: API.

Unified ELO 1715 ± 42, rank #301 of 1806 rated models, from 8 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
MATH-Perturb (Hard)88.89Accuracy (%)100
PlatinumBench (MIT)0.53Avg Error Rate (%)100
Gapminder AI Worldview93.1Correct Rate (%)97.1
EQ-Bench Creative Writing v31281.8Elo60.6
SciArena1009.7Elo Rating51.4
GosuEvals58.3Score (%)44.4
Reward Hacking Benchmark (RHB)4.6Exploit rate (%) (self-reported)41.7
SEAL - SWE-Bench Pro (Private Dataset)10.14Score15.4

Interactive version: theaggregate.ai/model?slug=gemini-2-5-pro-preview · How It Works · Data refreshed daily, snapshot 2026-08-07.