MiniMax open-sources MiniMax-01 with a 4M-token context

Background

MiniMax is a Chinese AI company. Its abab line of text models was proprietary, meaning the weights were not published.^[2] Before January 2025 the company had not released an open-weight series of this kind.

What happened

In January 2025 MiniMax released the MiniMax-01 series. It consisted of the language model MiniMax-Text-01 and the vision-language model MiniMax-VL-01.^[2] Weights for both were posted to Hugging Face on 12 January 2025, and the technical report, arXiv 2501.08313, was published two days later on 14 January 2025.^[3][1]

Hugging Face is the platform where researchers and companies share model weights. The technical report is the document that describes the model's design and results. An open-weight release makes the trained parameters available for download, so people can run the model on their own equipment, within the terms of its license. A proprietary model is usually used only through the company's own service.

Key details

MiniMax-Text-01 has 456B total parameters, with 45.9B activated per token.^[1] Its architecture is a hybrid. A softmax attention layer comes after every seven Lightning Attention layers, and the design is combined with mixture-of-experts components.^[1]

The most distinctive figure in the release is the context length. The model was trained with a context of up to 1 million tokens, and it can handle up to 4 million tokens at inference, which is the context length available when the model is used.^[1] The headline of the release refers to this 4-million-token figure.

The licensing arrangement is split. The weights use the MiniMax Model Agreement, while the code is MIT-licensed.^[1]

Impact

The release marks a change in how MiniMax shared its models. The open weights of MiniMax-01 made the architecture and results available for researchers to examine and run.^[1]

For the archive, the event sits at the start of MiniMax's open-model line. It was followed by the reasoning model MiniMax-M1 in June 2025.^[2]