Mistral AI Non-Production License (MNPL) Announcement

Mistral AI has introduced the Mistral AI Non-Production License (MNPL), a new licensing framework designed to permit non-commercial and research use of its technology while ensuring that commercial entities building businesses on its models do so sustainably. This move aims to balance the company's commitment to open AI principles with the necessity of growing its business and funding further research.

Purpose and Objectives of the MNPL

The Mistral AI Non-Production License (MNPL) is designed to address the tension between open access and commercial viability. Mistral AI states that while it supports developers building high-margin products using its models, it notes that such commercial success does not always contribute back to the company's research and independence.

According to the announcement, the MNPL serves two primary functions:

• Supporting Research and Innovation: The license allows developers to use Mistral AI technology for non-commercial purposes and to support research work.
• Ensuring Commercial Fairness: It establishes a framework where those who build a business based on Mistral AI's work must do so in a way that is fair and sustainable for all parties involved.

Implementation and Model Availability

Codestral is the first model released under the MNPL. This license is intended to strike a balance between the commitment to openness and the company's responsibility to grow its business.

Despite the introduction of the MNPL, Mistral AI has explicitly stated that it will continue to release models and code under the Apache 2.0 license. The company is now managing two distinct families of products released under these two different licensing schemes (Apache 2.0 and MNPL).

Commitment to Openness in AI

Mistral AI positions the MNPL as part of a broader effort to defend openness in AI. The company argues that openness is a catalyst for innovation and a safeguard for transparency and accountability. It further claims that the debate around openness is sometimes "instrumentalised to entrench the position of incumbent players" in the competitive AI industry.