one year on
xAI releases Grok-1 as open-source model under Apache 2.0 license
The 314B-parameter mixture-of-experts model, the largest open-source mixture-of-experts model to date, is released with code and weights as Musk criticizes OpenAI's closed-source model policy.
Elon Musk’s xAI released Grok-1 today, the largest open-source mixture-of-experts model to date — a 314-billion-parameter mixture-of-experts (MoE) model with 8 experts, 2 active per token. The weights and architecture are available under the permissive Apache 2.0 license.
Grok-1 is the raw base model from the pre-training phase that completed in October 2023. It has not yet been optimized for specific applications or tuned with human feedback (RLHF). The company provides no details on training data and does not provide any ethical or safety guidelines.
The release follows Musk’s lawsuit against OpenAI, in which he criticizes its closed-source model policy. However, the practical impact is tempered: the model has not yet been optimized for specific applications.
The release marks a milestone in open-weight AI — the largest open-source MoE model so far — but also underscores the gap between releasing a base model and delivering a usable product.
One year later — open only if you can handle spoilers
Grok-1 remains one of the largest open-weights models as of mid-2026, but its influence was overshadowed by Meta's Llama 3 later that year. The model saw limited adoption due to its size and lack of tuning, though it spurred further releases of large MoE models under open licenses.