AI/TLDR

Mistral Medium 3.1

Mistral's cost-efficient frontier-class multimodal model for coding, STEM, and enterprise

Overview

Mistral Medium 3.1 (API id mistral-medium-2508, version 25.08) is Mistral AI's frontier-class multimodal model, released on August 12, 2025. It is an incremental refresh of Mistral Medium 3 that adds improvements in reasoning, coding, STEM, and tone control while keeping the same cost and deployment profile. The model accepts text and image input and returns text, with a 128K-token context window.

Mistral positions the Medium line under the banner 'Medium is the new large': it claims Medium 3 performs at or above 90% of Claude Sonnet 3.7 on benchmarks across the board, surpasses leading open models such as Llama 4 Maverick, and does so at roughly 8x lower cost. Pricing is $0.40 per million input tokens and $2.00 per million output tokens. Unlike the later Mistral Medium 3.5, Medium 3.1 is proprietary and is not released as open weights, though Mistral offers hybrid, on-premises, and in-VPC deployment for enterprise customers (four GPUs and above).

Mistral Medium 3.1 is available through Mistral's La Plateforme API as well as third-party gateways such as OpenRouter and the Vercel AI Gateway. It has since been deprecated (scheduled deprecation 2026-05-22, retirement 2026-08-31) in favor of Mistral Medium 3.5, the open-weight successor. Mistral has not published a detailed per-benchmark (MMLU/GPQA/HumanEval) breakdown for this version, so this page reports only the specifications the company and major model trackers have verified.

Released2025-08-12
LicenseProprietary (commercial API; enterprise self-hosted/in-VPC deployment available)
WeightsAPI only
Context128K
ArchitectureMistral's proprietary multimodal Medium-class model (text + image input, text output). Released as an incremental update over Mistral Medium 3 with improvements in reasoning, coding, STEM, tone, and instruction-following. Designed for simple deployment, including self-hosted environments of four GPUs and above. Exact parameter count and architecture details are not publicly disclosed by Mistral AI.
ModalitiesText, Vision
StatusDeprecated (retiring 2026-08-31; succeeded by Mistral Medium 3.5)

Pricing

Input$0.40 / 1M tokens per 1M tokens
Output$2.00 / 1M tokens per 1M tokens

Pricing announced for the Mistral Medium 3 line and consistently reported for Medium 3.1 by Artificial Analysis, OpenRouter, and the Vercel AI Gateway.

Pricing source ↗

Strengths

  • Strong coding and STEM performance for its price tier, a focus area Mistral highlights for the Medium line
  • Cost-efficient: $0.40 input / $2.00 output per 1M tokens, roughly 8x cheaper than the frontier models Mistral benchmarks it against
  • Multimodal: handles both text and image (vision) input
  • 128K-token context window for long documents and codebases
  • Flexible deployment, including hybrid, on-premises, in-VPC, and self-hosted on four GPUs and above for enterprises
  • Improved tone, reasoning, and instruction-following over the original Mistral Medium 3

Best for

  • Coding assistance and code generation
  • STEM reasoning and technical question answering
  • Enterprise workflows requiring on-premises or in-VPC deployment
  • Document and image understanding (vision input)
  • Cost-sensitive production applications that need frontier-class quality at a mid-tier price
  • Multilingual chat and content generation

How to access

ProviderModel ID
Mistral AI (La Plateforme) ↗mistral-medium-2508
OpenRouter ↗mistralai/mistral-medium-3.1
Vercel AI Gateway ↗mistral/mistral-medium

Mistral Medium — every version

The full lineage of the Mistral Medium line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.

VersionReleasedContextLicense
Mistral Medium 3.5current2026-04-28MIT
Mistral Medium 3.12025-08-12Open weights
Mistral Medium 32025-05-07Open weights
Mistral Medium (2023)2023-12Proprietary

FAQ

When was Mistral Medium 3.1 released?

Mistral AI released Mistral Medium 3.1 on August 12, 2025. Its API identifier is mistral-medium-2508 (version 25.08).

Is Mistral Medium 3.1 open weights?

No. Mistral Medium 3.1 is proprietary and available via API; its weights were not publicly released. The later Mistral Medium 3.5 is the open-weight successor in this line. Enterprises can, however, run Medium 3.1 through hybrid, on-premises, or in-VPC deployments arranged with Mistral.

How much does Mistral Medium 3.1 cost?

It is priced at $0.40 per million input tokens and $2.00 per million output tokens, the same pricing Mistral announced for the Medium 3 line and roughly 8x cheaper than the frontier models Mistral compares it against.

What is the context window and what modalities does it support?

Mistral Medium 3.1 has a 128K-token context window. It is multimodal, accepting both text and image (vision) input and producing text output.