Mistral Medium 3, the Mistral AI model in numbers

What you need to know

Mistral AI released Mistral Medium 3 on May 7, 2025. It has no declared reasoning mode: it answers directly, without an extra billed thinking step. Its weights are not published, so it only runs remotely, through the Mistral AI API or partner hosts (6 listed on models.dev). That makes it more than a year old: successors may have shipped since.

We file it under budget models (24 models in this category on the site).

Measured performance

Epoch AI publishes 3 results for Mistral Medium 3 on our basket's benchmarks: too few to compute a reliable capability index.

• GPQA Diamond (PhD-level science questions): 59.5% — the best tracked model, Gemini 4 Argon, reaches 96.4%
• OTIS Mock AIME (competition mathematics problems): 32.2% — the best tracked model, GPT-5.5 Pro, reaches 100%
• Humanity's Last Exam (multidisciplinary expert questions): 4.5% — the best tracked model, GPT-6 Astra, reaches 54.8%

What it costs

Mistral Medium 3 costs $0.40 per million input tokens and $2 per million output tokens (median of 6 providers, checked on October 22, 2026).

In practice, an agent task that reads 20,000 tokens and writes 2,000 costs about $0.012, or $12 for 1,000 such tasks.

That is 1.8 times the median output price of budget models ($1.13).

To estimate your own budget, use the agent cost calculator.

Technical specifications

Its context window reaches 262,144 tokens, roughly 393 pages of text at 500 words per page: enough for a long report or several contracts at once. A single answer can run to 104,857 tokens.

It accepts as input: text, image and PDF. Its knowledge stops at March 31, 2025: beyond that, it needs web search or your documents.

It can call tools (functions, APIs) and return structured JSON output, which makes it a possible foundation for an AI agent.

When to choose it

Given its category, Mistral Medium 3 is mainly aimed at high volumes where every cent counts: sorting, information extraction, summaries, simple assistants.

Before adopting it, test it on your own examples: benchmarks measure standard tasks, not your job.

What could replace it?

In the same category, the models closest to Mistral Medium 3 are:

• GPT-5 Mini (OpenAI, index 62.9)
• Grok Build 0.1 (xAI, $2 per million output tokens)
• Qwen3.7 Plus (Alibaba Cloud (Qwen), index 56.2)
• Gemini 3.1 Flash Lite (Google DeepMind, index 64.3)