Mixtral 8x7B
Mixtral 8x7B is Mistral’s first public mixture-of-experts family. It provides Base and Instruct checkpoints with a 32K context and support for five European languages and code.
Instruction training and deployment
The instruction version combines supervised fine-tuning with direct preference optimization. The release describes English, French, German, Italian and Spanish. At launch, the hosted mistral-small endpoint used Mixtral; that historical endpoint label is not the later proprietary Small family.Mixtral 8x7B release announcement
One family, two checkpoint roles
Mixtral-8x7B-v0.1 is the pretrained model. Mixtral-8x7B-Instruct-v0.1 adds instruction training and a chat template. Their cards link the original torrent release. The family date uses December 2023 because the base release preceded the dated instruction/API announcement.Mixtral 8x7B official model cardMixtral 8x7B primary source
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | December 2023 |
| Access | Downloadable |
| Model identifiers | Mixtral-8x7B-v0.1Mixtral-8x7B-Instruct-v0.1open-mixtral-8x7b |
| Context window | 32,000 tokens The model catalog advertises 32k context for this version. This is not a measured recall limit. Mixtral 8x7B model and lifecycle record |
| Weight license | Apache 2.0Mixtral 8x7B official model card |
| Access and lifecycle | Hosted API retired on 30 March 2025. Downloadable weights remain listed.Mixtral 8x7B model and lifecycle recordMixtral 8x7B official model card |
| Checkpoint variants | Mixtral-8x7B-v0.1 and Mixtral-8x7B-Instruct-v0.1Mixtral 8x7B official model cardMixtral 8x7B primary source |