Exciting news, Grok-1 is now available and fully open-sourced. This set a new standard for AI transparency and collaboration.

In a landmark move that has sent shockwaves through the AI community, xAI has open-sourced its state-of-the-art language model, Grok-1.

This 314 billion parameter Mixture-of-Experts model, trained from scratch by xAI, represents a significant step towards making advanced AI technology more accessible and transparent.

MoE architecture

MoE architecture

The release of Grok-1 has been met with widespread excitement and praise, with many experts hailing it as a pivotal moment in the journey towards open-source Artificial General Intelligence (AGI).

The Grok-1 model

Grok-1 is a raw base model checkpoint from the pre-training phase, which concluded in October 2023. It has not undergone any fine-tuning for specific applications, such as dialogue.

The model boasts an impressive 314 billion parameters and employs a Mixture-of-Experts architecture, with 25% of the weights active on a given token.

xAI trained Grok-1 from scratch using a custom training stack built on top of JAX and Rust, showcasing their commitment to pushing the boundaries of AI development.

To start working with the model, interested parties can follow the detailed instructions provided on xAI’s GitHub repository: github.com/xai-org/grok. The repository contains example code for loading and running the Grok-1 open-weights model, along with the necessary requirements and dependencies.

More details about the architecture can be found here: https://x.ai/blog/grok other link for the weights: https://academictorrents.com/details/5f96d43576e3d386c9ba65b883210a393b68210e

And yes, license is Apache 2.0.

Performance and benchmarks

Grok-1 Shines on Complex Tasks Grok-1 has demonstrated remarkable performance on a range of standard machine learning benchmarks designed to measure math and reasoning abilities.

It outperformed all other models in its compute class, including ChatGPT-3.5 and Inflection-1, on challenging tasks such as GSM8k (middle school math word problems), MMLU (multidisciplinary multiple-choice questions), HumanEval (Python code completion), and MATH (middle school and high school mathematics problems written in LaTeX).

Grok 1 Benchmark

Grok 1 benchmarks

Grok-1 was only surpassed by models trained with significantly larger amounts of data and compute resources, like GPT-4, highlighting its efficiency and potential.

In a real-world test on the 2023 Hungarian national high school finals in mathematics, Grok-1 achieved a C grade (59%), demonstrating its ability to handle complex, unseen problems.

This impressive performance showcases the model’s potential for tackling real-world challenges and its ability to generalize to new domains.

The significance of open-sourcing Grok-1

A significant shift in the AI landscape occurred when xAI’s and Elon Musk decided to open-source Grok-1, setting a new standard for transparency and collaboration in the field of AI.

By making the model weights and architecture freely available under the Apache 2.0 license, xAI has opened the door for researchers, developers, and organizations worldwide to access, study, and build upon a state-of-the-art language model.

The implications of this move are far-reaching:

  1. Accelerated innovation: With access to Grok-1, the AI community can now collaborate and innovate at an unprecedented pace, exploring new applications, optimizations, and extensions of the model.
  2. Enhanced transparency: Open-sourcing Grok-1 promotes greater understanding of the model’s inner workings, fostering trust and accountability in AI systems. This transparency is crucial for addressing concerns around bias, fairness, and ethical considerations in AI development.
  3. Democratizing AI: By making Grok-1 freely available, xAI has taken a significant step towards democratizing access to advanced AI technology, enabling a wider range of individuals and organizations to participate in the development and application of cutting-edge language models.

Challenges and future work

While the release of Grok-1 is a groundbreaking achievement, there are still challenges to address. The model’s large size makes it difficult for the open-source community to iterate on it directly, requiring significant computational resources. However, experts anticipate that functional quantized versions will be available within the next month, making Grok-1 more accessible to a broader range of researchers and developers.

Companies like Abacus AI have already expressed their commitment to fine-tuning Grok-1, with updates and releases expected in the coming weeks. This collaborative effort highlights the potential for rapid progress and innovation as the AI community works together to unlock the full potential of Grok-1.

Looking ahead

xAI’s Vision for the Future of AI The open-sourcing of Grok-1 is just the beginning of xAI’s ambitious roadmap for advancing AI technology. As outlined in their blog post, xAI is dedicated to developing AI tools that assist humanity in its quest for understanding and knowledge.

They aim to gather feedback, ensure broad accessibility, and empower research and innovation through their AI tools.

xAI has identified several key research directions that will shape the future of Grok and AI as a whole. These include scalable oversight with tool assistance, integrating formal verification for safety and reliability, improving long-context understanding and retrieval, enhancing adversarial robustness, and incorporating multimodal capabilities. By focusing on these areas, xAI aims to create AI systems that can reason deeply about the real world, provide formal guarantees for code correctness, and contribute significant scientific and economic value to society.


The release of Grok-1 marks the beginning of a new era in open-source AI, one characterized by transparency, collaboration, and rapid innovation. By making Grok-1’s weights and architecture freely available, xAI has set a new standard for the AI community, demonstrating their commitment to advancing the field responsibly and inclusively.

As researchers, developers, and organizations worldwide begin to work with and build upon Grok-1, we can expect to see a surge in breakthroughs and applications that push the boundaries of what is possible with AI. The future of AI is looking brighter than ever, and xAI’s Grok-1 is leading the charge towards a more open, accessible, and impactful AI ecosystem.

With the support and collaboration of the global AI community, Grok-1 has the potential to accelerate our progress towards open-source AGI, unlocking new possibilities for science, technology, and society as a whole. As we embark on this exciting journey, we can look forward to a future where AI is a powerful tool for empowerment, understanding, and positive change.


Last Update: 18/03/2024