Mistral AI Launches Open-Source 7B Parameter Language Model Ahead of Schedule
Paris-based startup Mistral AI releases Mistral 7B, a highly capable open-source language model available for commercial use under an Apache 2.0 license. The new model outperforms competing open models up to 13 billion parameters on standard benchmarks.
Well-funded artificial intelligence startup Mistral AI releases an open-source language model called Mistral 7B, which features 7 billion parameters and is available on GitHub under an Apache 2.0 license. The Paris-based company, founded just five months ago by former researchers from Meta and Google, allows users to utilize this model for both research and commercial applications.
The startup completes the development of Mistral 7B in only three months, a timeline that exceeds the company's original expectation to introduce its first language models in 2024. During this rapid development cycle, the team assembles an engineering group and builds a complete MLOps stack to support neural network creation. The model handles various text processing tasks, including prose generation, document summarization, and software code autocompletion.
Mistral AI claims that Mistral 7B outperforms all currently available open models up to 13 billion parameters on standard English and code benchmarks, effectively beating Meta's Llama 2 13B. The model supports a context length of 8,000 tokens per prompt and even achieves performance on par with much larger 34 billion parameter systems, marking a significant milestone for the young company.