Mistral releases Medium 3.5, a dense 128B model with a modified MIT licence
Mistral Medium 3.5 is a dense 128B instruct, reasoning and coding model with 256k context. The licence excludes large-revenue companies.
Mistral AI published Mistral Medium 3.5 on Hugging Face. The card calls it the company's first flagship merged model: a dense 128B model with a 256k context window that handles instruction following, reasoning and coding in one set of weights. It replaces Mistral Medium 3.1 and Magistral in Le Chat and Devstral 2 in the Vibe coding agent.
Input is text and images, output is text. Reasoning effort is set per request, with none or high as the recommended values. The card lists dozens of languages. A separate EAGLE draft model is provided for faster local serving with vLLM or SGLang. The card also warns that an earlier Transformers config had an incorrect entry that degraded long-context performance, since fixed in a commit.
The licence is a Modified MIT License: open for commercial and non-commercial use, with exceptions for companies with large revenue.
The card gives no release date; the date used is the Hugging Face repo creation date.
Open unless you are large. Read the exception.
Sources
Editor's first pass (AI-assisted)
- Real 4: Ungated weights with an EAGLE draft model; the modified MIT licence has revenue exceptions.
- Useful 4: One model for chat, reasoning and coding, with configurable reasoning effort.
- Notable 3.5: Replaces Medium 3.1, Magistral and Devstral 2 in Mistral's own products, per the card.
- Novel 3: A merge of previous lines into one set of weights.
Comments