Grok 2
Release in the Grok family · version Grok 2
Maintained by xAI (SpaceXAI)12
Grok 2 is a post-trained language model that xAI trained and used in 2024; its weights, about 500 GB in 42 files, were published on Hugging Face in August 2025. Press coverage at the time reported the release as Grok 2.5.13
- Model hub: Hugging Face model repository (external site: huggingface.co)
- License: LICENSE (xAI Community License Agreement) (external site: huggingface.co)
Availability and license
Overall availability
Downloadable from Hugging Face without gating; use is governed by the xAI Community License Agreement, which conditions commercial use on following xAI's Acceptable Use Policy and bars using the materials or outputs to train other general-purpose models.12
Availability is separate from permission: read the license before using or redistributing.
The model card calls it the "Grok 2 Community License Agreement", but the LICENSE file is titled xAI Community License Agreement (last updated November 4, 2025). It grants a revocable license for non-commercial and research use, and allows commercial use only if users follow xAI's Acceptable Use Policy. It forbids using the materials or their outputs to train or improve other foundation, large language, or general-purpose models (fine-tuning Grok 2 itself is allowed), requires "Powered by xAI" attribution when distributing, ends if the licensee brings certain IP lawsuits, and is governed by Texas law.21
Model-disclosure tier
The model parameters for this release can be downloaded by the public. License terms may still restrict use, redistribution, or commercial use.
The weights are under a license that is not on the rubric's OSI-approved list. Read its terms before use.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| WeightsCan the general public download the model parameters for this release? | Public | Public download under a custom license with use restrictions.12 |
| Inference codeIs code for running the model published? | Public | Serving is documented through the open-source SGLang engine (v0.5.1 or later); the xAI repository holds weights, config, and tokenizer but no inference code of its own.1 |
| Training codeIs the code used to train the model published? | Unknown | Not assessed. |
| Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access. | Unknown | Not assessed. |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Unknown | Not assessed. |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Unknown | Not assessed. |
Run and use notes
- The model card documents serving with SGLang using tensor parallelism across 8 GPUs, each with more than 40GB of memory; the documented launch command uses fp8 quantization and the Triton attention backend.1
- The model card says this is a post-trained model and that requests need the correct chat template.1
Organization context
Other releases in the Grok family
- Grok-1Model-disclosure tier (USASI rubric v0.1): Open-weight
U.S. eligibility
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.