Mistral Medium 3.5 sits in the mid-tier of the Mistral family and is positioned as a dense 128-billion-parameter instruction-following model. According to the Requesty listing, it accepts text and image inputs while returning text, with a custom vision encoder that handles variable image sizes and aspect ratios rather than being limited to fixed crops. The model's design emphasizes agentic workflows, coding, and complex multi-step reasoning, and it is highlighted as particularly strong at reliable multi-tool calling and long-horizon tasks. Configurable reasoning effort per request lets callers dial the depth of deliberation up or down depending on the workload.
Practically, the Requesty route exposes a long context window that the vendor describes as 256K tokens, well suited to retrieval-augmented agents, codebases, and document-heavy pipelines that need to keep large amounts of material in working memory. The same page notes the model can be self-hosted on as few as four GPUs, suggesting an accessible inference path for teams that want on-prem control alongside the managed EU endpoint. For builders, this makes the model a reasonable pick when the priority is sustained tool use and step-by-step problem solving across long contexts, rather than pure conversational chat, while still benefiting from multimodal input for grounding on diagrams, screenshots, or document images.
