Mistral: Mistral Medium 3.1 (batch)

Mistral Medium 3.1 is Mistral AI's August 2025 refresh of Mistral Medium 3, positioned as a frontier-class multimodal enterprise model that accepts text and image inputs and returns text. It targets professional workloads including coding, STEM reasoning, document Q&A, function calling, structured outputs, and agent-style workflows, with a 128k-token context window and a June 2025 knowledge cutoff. Mistral describes the release as improving overall reasoning, coding, and multimodal quality plus more consistent tone and stronger web-search behavior compared with its predecessor.

The batch offering (API id mistral-medium-2508 via Mistral's /v1/batch endpoint) is the same underlying model served for asynchronous bulk jobs at reduced per-token pricing relative to the standard chat API. It supports the same modalities and tooling as the realtime endpoint and is suited to large offline inference pipelines such as enrichment, classification, and batch content generation. Weights are proprietary and hosted only through Mistral's cloud and partner platforms.

Independent evaluations from Artificial Analysis report strong non-reasoning performance for its price tier, with competitive scores on MMLU-Pro, GPQA Diamond, LiveCodeBench, and AIME 2025, while agentic terminal benchmarks remain modest. The model is deprecated in Mistral documentation in favor of Mistral Medium 3.5 for new integrations, with a documented deprecation date of May 22, 2026.