Atlas

Models

← All models

Grok 3 Think (source unspecified)

xAI

This generative benchmark entry preserves results explicitly identified as Grok 3 (Think) or Grok 3 Beta Extended Thinking. xAI documents a separately trained consumer reasoning model, but the benchmark sources do not identify an immutable checkpoint or canonical API identifier. These observations are separate from the non-reasoning grok-3 API artifact.

Benchmark scores

BenchmarkScore
AIME 202577.3
GPQA Diamond80.2
LiveCodeBench - October 2024 to February 202570.6
MMMU76.0
MRCR v2 (8-needle) - 128K average34.0
USAMO 20254.8
Loading Atlas data…