Atlas

Models

← All models

Grok 3 Mini

xAI · 2025-02-19

Grok 3 Mini is xAI's lightweight thinking model released in February 2025, designed for cost-efficient reasoning on STEM tasks including mathematics, logic puzzles, and coding challenges. It operates as a chain-of-thought model that reasons step by step before producing a final answer, exposing its reasoning trace for transparency, and achieved strong results on math benchmarks relative to its size. The model carries a 131,072-token context window and is best suited for structured reasoning tasks that do not require broad world knowledge.

Benchmark scores

BenchmarkScore
AA-LCR50.3
AA-Omniscience Index-6.0
AidanBench2285
Aider Polyglot49.3
AIME 2024 (avg@8)88.8
AIME 2024-202585.0
AIME 202565.0
AIME 2025 (avg@8)81.3
ARC-AGI-116.5
ARC-AGI-20.4
Arena Text — Business, Management & Financial Ops (No Style Control)1345
Arena Text — Chinese (No Style Control)1379
Arena Text — Creative Writing (No Style Control)1340
Arena Text — English (No Style Control)1377
Arena Text — Entertainment, Sports & Media (No Style Control)1333
Arena Text — Exclude Ties (No Style Control)1332
Arena Text — Exclude Ties (Style Controlled)1326
Arena Text — Expert (No Style Control)1390
Arena Text — French (No Style Control)1355
Arena Text — French (Style Controlled)1356
Arena Text — German (No Style Control)1354
Arena Text — Hard Prompts (No Style Control)1373
Arena Text — Hard Prompts English (No Style Control)1378
Arena Text — Instruction Following (No Style Control)1357
Arena Text — Japanese (No Style Control)1338
Arena Text — Korean (No Style Control)1342
Arena Text — Legal & Government (No Style Control)1386
Arena Text — Life, Physical & Social Science (No Style Control)1379
Arena Text — Longer Query (No Style Control)1372
Arena Text — Medicine & Healthcare (No Style Control)1391
Loading Atlas data…