Mistral released the weights of Mistral Large 4 after the hosted preview. The model combines image and text input with a sparse expert architecture and a million-token context window, giving teams a self-hosted option for workloads that use long documents and images.
Mistral AI
Also known as: Mistral · Mistral 7B · Mistral Large · Mixtral · Ministral
Plain English
Mistral AI is a French model developer that publishes language and multimodal models. Developers can access its models through hosted services and download weights for selected releases.
Technical Overview
Its model families span dense models such as Mistral 7B and sparse mixture-of-experts models such as Mixtral and the later Large releases. Several generations include base and instruction-tuned Checkpoints, giving developers a choice of starting points for adaptation or deployment.
Chronological Entries (4)
Mistral released Mistral Large 3 alongside 3B, 8B, and 14B Ministral models under Apache 2.0. The smaller models included base, instruction-tuned, and reasoning variants with image input support, offering several starting points for local deployment or adaptation.
Mistral released Mixtral 8×7B with Apache 2.0 weights. Its expert router selects two of eight feed-forward experts per token in each layer, so a token uses only a fraction of the model’s parameters.
Mistral released Mistral 7B under Apache 2.0. The model uses grouped-query attention, which lets several query heads share key and value heads and reduces the amount of data held in the KV cache.
Direct Relations (1)
- Part ofEuropean AI Labs