Mistral 7B

A French startup released a seven-billion-parameter model that outperformed ones twice its size, under an Apache licence with no restrictions.

Why it matters

Europe acquired a frontier player of its own, and a model released under an Apache 2.0 licence with no restriction on its use.

The company was founded a few months earlier by people from Meta and DeepMind. Unlike Llama 2, the Apache 2.0 licence restricts neither the size of the user nor the field of application. That made Mistral the basis for products where Meta's licence was unacceptable. The arrival of a European player coincided with the passage of European AI regulation. It was not the first such release: MosaicML's MPT-7B (5 May 2023) was licensed Apache-2.0 for commercial use, and TII's Falcon-40B model card also states Apache 2.0 with commercial use permitted.

Event record

Event date

September 27, 2023

Timeline date

Event date

Verification

Sources gathered automatically · September 17, 2026

Lines

Openness

ID

evt-0282

The model was announced on 27 September 2023.

Sources

primary Announcing Mistral 7B
Mistral AI · Published September 2023
primary Introducing MPT-7B: A New Standard for Open-Source, Commercially Usable LLMs
Databricks (MosaicML blog) · Published May 5, 2023
primary tiiuae/falcon-40b model card
Technology Innovation Institute, via Hugging Face · Published May 24, 2023

Related events

Earlier

Related Llama 2 2023
Weights under an Apache licence with no restrictions on use.
Related The draft European AI Act 2021
A European frontier player appeared alongside European regulation.

Later

Extends Mixtral 8x7B 2023links to this record
The same company moved to a mixture of experts two and a half months later.
- Mixtral of experts · Mistral AI
Builds on A first exam for open models in Ukrainian 2024links to this record
The report lists Mistral 7B among the allowed open models, and the best team, Sherlock, fine-tuned Mistral-7B-Instruct.
- The UNLP 2024 Shared Task on Fine-Tuning Large Language Models for Ukrainian · ELRA and ICCL (Proceedings of the Third Ukrainian Natural Language Processing Workshop @ LREC-COLING 2024, pages 67-74)