The Open-Source AI Stack
RSS
All models

Models · grok

Grok 3

Proprietary xAI · 2025-02-17 · Proprietary

xAI's third-generation flagship, trained on the Colossus supercomputer (approximately 200,000 GPUs) with roughly 10x the compute of Grok 2. Released alongside a separate Grok 3 Reasoning variant and a DeepSearch product, with xAI claiming wins over GPT-4o on AIME math and GPQA science benchmarks. API access launched in April 2025.

Cost

$4.00 / Mtok input
$20.00 / Mtok output

xAI API · as of 2026-05-21

source ↗

Speed

0 tok/sec output
0 ms TTFT

· as of 2026-05-21

source ↗

Architecture

tokens in Embedding vocab not disclosed × N layers Architecture not disclosed (proprietary or undocumented) Output projection tokens out
Schema-generated from data/models.yaml. Every label is auditable against the model's sources.

Specs

Architecture
unknown
Total params
not disclosed
Active params
not disclosed
Context window
131K tokens
Attention
unknown
Position encoding
unknown
Training hardware
H100
Post-training
rlhf
OSI-approved
no
Data released
no
Training code
not released

Benchmarks

Each score carries the date it was published; we never infer or interpolate missing scores.

General reasoning

MMLU-Pro 79.9 as of 2026-05-21 source ↗
GPQA-Diamond 69.3 as of 2026-05-21 source ↗

Code

LiveCodeBench 42.5 as of 2026-05-21 source ↗

Math

MATH 87.0 as of 2026-05-21 source ↗
AIME 2024 33.0 as of 2026-05-21 source ↗
AIME 2025 58.0 as of 2026-05-21 source ↗

Available quantizations

None. The weights are not distributed, so there are no public quantizations.

Notable innovations

  • · Think mode and Big Brain mode for reasoning
  • · DeepSearch over web plus X
  • · Trained on Colossus 200K GPU cluster

Sources