Robots Atlas>ROBOTS ATLAS
Grok-2

Grok-2

2ย ยทย Family: Grok
xAI second-generation flagship model with multimodal capabilities and image generation via FLUX by Black Forest Labs. Weights available on HuggingFace (~500 GB, 42 files). Requires 8ร— GPU with >40 GB memory each.
โš  Deprecatedโœ“ Public accessโš– Open weightsLLMMultimodal๐Ÿ“ Grok
Context window
131K
tokens
Parameters
nieujawnione
parameters
Release date
20 August 2024
Access:APIDownloadHostedDeployment:โ˜ Cloud๐Ÿ’ป Local

Overview

Grok-2 is xAI's multimodal frontier language model announced on 13 August 2024 and rolled out to X Premium and Premium+ subscribers. In xAI's official August 2024 benchmarks it achieved GPQA 56.0%, MMLU 87.5%, MMLU-Pro 75.5%, MATH 76.1%, HumanEval 88.4%, MMMU 66.1%, MathVista 69.0%, and DocVQA 93.6%. An early version tested on the LMSYS Chatbot Arena under the codename "sus-column-r" outperformed Claude 3.5 Sonnet and GPT-4-Turbo on overall Elo at the time. The model integrates image generation through a Black Forest Labs FLUX.1 partnership. In August 2025 xAI released the Grok-2 weights on Hugging Face under the xAI Community License Agreement (source-available, with commercial-use restrictions) โ€” the checkpoint is ~500 GB across 42 files and requires 8 GPUs with >40 GB each (TP=8, FP8 quantization). The exact parameter count has not been officially disclosed by xAI.

Classification
LLMMultimodal
Family: Grok
Access & deployment
APIDownloadHosted
CloudLocal
Weights: Open weights
Key parameters
๐Ÿ“ Context: 131K
๐Ÿงฉ Parameters: nieujawnione
๐Ÿ“ฅ Input: text, image

Technical specification

Context window
131K
tokens
Parameters
nieujawnione
parameters
License
xAI Community License Agreement
Modalities
โฌ‡ Input
textimage
โฌ† Output
textimage

Capabilities and applications

Native model capabilities
Reasoning
The model's ability to reason logically and solve complex problems.
Category: reasoning
Coding
Generating, analysing and modifying source code.
Category: coding
Image understanding
Analysing and interpreting the content of images.
Category: vision
Multilingual
Understanding and generating text in many languages.
Category: language
Multi-step reasoning
Carrying out multi-step chains of reasoning across long, complex tasks.
Category: reasoning

Benchmark results

8 benchmarks
GPQA
0-shot CoT (xAI eval, Aug 2024)
56.0%
๐Ÿ“„ xAI Grok-2 Beta Release blog
MMLU
0-shot CoT (xAI eval, Aug 2024)
87.5%
๐Ÿ“„ xAI Grok-2 Beta Release blog
MMLU-Pro
0-shot CoT (xAI eval, Aug 2024)
75.5%
๐Ÿ“„ xAI Grok-2 Beta Release blog
MATH
maj@1 (xAI eval, Aug 2024)
76.1%
๐Ÿ“„ xAI Grok-2 Beta Release blog
HumanEval
pass@1 (xAI eval, Aug 2024)
88.4%
๐Ÿ“„ xAI Grok-2 Beta Release blog
MMMU
0-shot CoT (xAI eval, Aug 2024)
66.1%
๐Ÿ“„ xAI Grok-2 Beta Release blog
MathVista
xAI eval, Aug 2024
69.0%
๐Ÿ“„ xAI Grok-2 Beta Release blog
DocVQA
xAI eval, Aug 2024
93.6%
๐Ÿ“„ xAI Grok-2 Beta Release blog

Technical architecture