Mistral has released the weights for Mistral Large 4, moving beyond its earlier hosted preview. The original checkpoint history records an initial release on October 29 at 14:11 UTC. Our coverage is dated October 29.
What is available
The official model card describes a multimodal mixture-of-experts model with roughly one trillion total parameters. It lists 49 billion active parameters per token, or 52 billion when embeddings and output layers are included, and a one-million-token context window. It accepts text and images and generates text.
Mistral documents vLLM and SGLang serving, an embedded DSpark draft head for speculative decoding, and a separate NVFP4 checkpoint targeting Blackwell GPUs. These details do not establish support in our launcher or a practical fit for a 12 GB RTX 3060.
Open weights, with a specific business condition
The Mistral Ecosystem Open-Weight License is a modified MIT license. If a licensee or its affiliates operates a Model-as-a-Service business, a separate commercial agreement is required before commercial use. Its definition distinguishes that business from embedded end-user features and mere forwarding to another host, and lists internal-use and official-product exceptions. Read the actual conditions before building a hosted offering; calling this simply MIT would omit a material restriction.
What the results establish
Mistral reports coding, security and enterprise benchmark improvements. Those remain attributed results, not an AiBrainWorX evaluation. Downloadable weights make independent deployment possible; they do not make every deployment inexpensive or every generated patch correct. We have not downloaded the weights, tested throughput or changed our published launcher benchmark.
Original AiBrainWorX reporting. Sources checked October 29, 2026. No model weights were downloaded and no independent performance test was performed.
