Report
Mistral 7B: a 7B model that outperforms Llama 2 13B, released under Apache 2.0
Mistral AI has released Mistral 7B, a 7.3B parameter language model under the Apache 2.0 license. The company says it outperforms Llama 2 13B on all benchmarks and can be used without restrictions, including locally.
Mistral AI announced the release of Mistral 7B on September 27, 2023. According to the announcement, it is a 7.3B parameter model that outperforms Llama 2 13B on all benchmarks and Llama 1 34B on many benchmarks. It also approaches CodeLlama 7B performance on code while remaining good at English tasks. The model uses Grouped-query attention (GQA) for faster inference and Sliding Window Attention (SWA) to handle longer sequences at smaller cost.
The practical difference for users is the license and deployment options. Mistral 7B is released under the Apache 2.0 license, which the announcement says allows it to be used without restrictions. It can be downloaded and used anywhere, including locally, with the reference implementation. It can be deployed on any cloud (AWS, GCP, Azure) using the vLLM inference server and skypilot, or used on HuggingFace. The announcement also states that it is easy to fine-tune on any task, and a chat fine-tuned version is provided that outperforms Llama 2 13B chat.
The source claims that Mistral 7B is "the most powerful language model for its size to date." It also says the company re-ran all model evaluations themselves for fair comparison. For someone choosing a model for everyday work, the combination of a small footprint, permissive license, and local deployment could make it a practical alternative to larger models that require more resources or have more restrictive terms.
However, the evidence here is limited to Mistral AI's own announcement. The benchmarks are reported by the company, and the announcement does not include independent third-party evaluations. Readers should treat the performance claims as the company's own until verified elsewhere. The key takeaway is that a 7B model is now available under Apache 2.0 with claimed performance beyond its size class, which may lower the barrier to running capable language models locally or on modest cloud instances.
Source details and supporting facts
Each line is stated by the page named above it.
Stated by mistral.ai
- Mistral 7B is a 7.3B parameter model.
- Mistral 7B outperforms Llama 2 13B on all benchmarks.
- Mistral 7B is released under the Apache 2.0 license.
- Mistral 7B can be used without restrictions.
- Mistral 7B uses Grouped-query attention (GQA) for faster inference.
- Mistral 7B uses Sliding Window Attention (SWA) to handle longer sequences at smaller cost.
- Mistral 7B can be downloaded and used locally with the reference implementation.
- Mistral 7B can be deployed on any cloud (AWS/GCP/Azure) using vLLM inference server and skypilot.
- Mistral 7B can be used on HuggingFace.
- Mistral 7B is easy to fine-tune on any task.
- A chat fine-tuned version of Mistral 7B outperforms Llama 2 13B chat.
- Mistral 7B approaches CodeLlama 7B performance on code, while remaining good at English tasks.
- Mistral 7B outperforms Llama 1 34B on many benchmarks.
- The announcement was published on 2023-09-27.
Sources
- Mistral AI newsText stored 15 September 2026
How this story was checked. Written from the 1 page listed above, stored 15 September 2026; claims checked against that stored text on 15 September 2026.
What that means
- 14 of 14 reported statements were confirmed against the page that carries them; the rest were removed rather than published.
- Figures in the text were required to appear in the stored source text: yes. Identifiers: yes.
- The check reads stored text only: no claim rests on a fresh look that did not happen.
- Where the reporting was silent, the text says so instead of filling the gap.