AI Flash Report

AI benchmarks

Sortable leaderboards for the benchmarks frontier labs report. Pick one to see every model's score.

GPQA Diamond

40 models scored

Leader: MiniMax-M3 92.9%

MMLU-Pro

36 models scored

Leader: Gemini 3.1 Pro 93.8%

HLE

36 models scored

Leader: Claude Fable 5 53.3%

TAU2-bench

27 models scored

Leader: JT-35B-Flash 99.1%

LiveCodeBench

38 models scored

Leader: GPT-5.2 88.9%

SciCode

35 models scored

Leader: Claude Fable 5 60.2%

ARC-AGI-2

1 models scored

Leader: Gemini 3.1 Pro 77.1%

SWE-bench Verified

14 models scored

Leader: GPT-5.3 Codex 82.4%

HumanEval

26 models scored

Leader: GPT-5.3 Codex 96.8%