PlatinumBench (MIT) — leaderboard

Carefully curated benchmarks testing LLM reliability - manually revised to remove dataset errors. Measures consistency and correctness across math, logic, reading comprehension, and commonsense tasks.

Metric: Avg Error Rate (%). Source: platinum-bench.csail.mit.edu. Status: saturated. 34 models tracked.

Top models

#ModelScore
1O1 (High)0.74
2Grok 30.79
3GPT-4.5 (Preview)0.99
4Claude 3.7 Sonnet1.05
5Claude 3.5 Sonnet1.09
6GPT-4.1 (2025-04-14)1.11
7O1 (Medium)1.27
8DeepSeek R11.4
9Grok 3 Mini (High)1.51
10Claude 3.5 Sonnet (20240620)1.84
11Llama 3.1 405B Instruct1.92
12Qwen 2.5 Max2.05
13Llama 4 Maverick2.12
14GPT-4o (2024-08-06)2.24
15O1 Preview2.33

Interactive version: theaggregate.ai/benchmark?slug=platinumbench-mit · How the rankings work · Data refreshed daily, snapshot 2026-07-22.