Atlas

Models

← All models

Grok 3 (API)

xAI · 2025-04-03

Grok 3 is xAI's non-reasoning API model, publicly available from April 3, 2025, with text input and output and a 131,072-token context window. xAI lists grok-3-beta and the grok-3-fast aliases under this same API model; faster serving is an inference configuration. The consumer Grok 3 Think model and earlier preview results are separate identities.

Benchmark scores

BenchmarkScore
A-Fantasia47.0
A-Fantasia (Valid Attempts with Retries)47.1
A-Fantasia Backwards Spelling36.0
A-Fantasia Chess28.0
A-Fantasia Chess (Valid Attempts with Retries)28.0
A-Fantasia Cube (Valid Attempts with Retries)77.0
A-Fantasia Cube Rotation77.0
A-Fantasia Spelling (Valid Attempts with Retries)36.4
AA-LCR54.7
AA-LCR v1.158.0
AA-Omniscience Index-33.2
AidanBench1796
Aider Polyglot53.3
AIME 2024 (avg@8)60.4
AIME 2024-202558.8
AIME 2025 (Artificial Analysis avg@10)58.0
AIME 2025 (avg@8)57.1
ARC-AGI-1 Semi-Private5.5
ARC-AGI-20.0
Artificial Analysis Omniscience Accuracy28.9
Artificial Analysis Omniscience Hallucination Rate87.3
Artificial Analysis Progress Index v1194
BALROG29.5
BALROG — BabaIsAI (LLM)33.3
BALROG — BabyAI (LLM)62.0
BALROG — Crafter (LLM)25.0
BALROG — MiniHack (LLM)22.5
BALROG — NetHack (LLM)1.6
BALROG — TextWorld (LLM)32.5
Capability136.8
Loading Atlas data…