0

Grok 3

Closed

xAI's Feb 2025 reasoning-capable frontier model, trained on the 200K-GPU Colossus cluster, with "Think" mode and a 1M-token context.

Open
Publisher
xAI
Family
Grok 3
Params
Undisclosed (closed)
Context
1M
Input
Output
Providers
xai
License
Proprietary
Released
Feb 2025
Canonical
x.ai/news/grok-3

Cite

Notes

Only stored in your browser.

Attribution

Benchmark scores
AAfrontiermathLiveBenchVals AI
Attribution policy →
Context
1M
Input
$4
/1M tok
Output
$20
/1M tok
Blended
$8
#398/ 414/1M · 3:1
Speed
0
#221/ 414tok/s

Price & speed via Artificial Analysis · ranked across tracked models

Reported on 17 evals across 8 domains - best 87.0% on MATH-500 · #65 of 178

Reported eval scores

17
TaxEval v275.9
CorpFin v259.7
MMLU-Pro79.9
Humanity's Last Exam (HLE)5.1
LiveBench52.7
SciCode36.8
GPQA Diamond69.3
AIME 2024: Problems from the American Invitational Mathematics Examination33
MATH-50087
FrontierMath3.8
IFBench46.9
LiveBench - Language51.4
Terminal-Bench (Hard)11.4
LiveBench - Coding53.9
AIME 2025: Problems from the American Invitational Mathematics Examination58
LiveCodeBench42.5
τ²-bench (Tau²-bench)48.8