# Magistral 1.0 (Small & Medium)

> Mistral AI's first reasoning models — a 24B Apache-2.0 open-weight Small and an API-only Medium, both built for transparent, multilingual chain-of-thought.

- Maker: Mistral AI
- Line: Magistral
- Canonical page: https://ai-tldr.dev/models/magistral-1-0/
- Updated: 2026-06-18

## In short

Magistral 1.0 (Small & Medium) is Mistral AI's first reasoning models — a 24B Apache-2.0 open-weight Small and an API-only Medium, both built for transparent, multilingual chain-of-thought. Mistral AI released it on 10 June 2025. It has a 128K-token context window. You can use it through Mistral La Plateforme, Hugging Face (self-host) and OpenRouter.

## Facts

| Fact | Value | Source |
| --- | --- | --- |
| Context window | 128K (max output Not separately published; recommended working window is 40K tokens (performance may degrade beyond that, within the 128K maximum).) | [1] |
| Input price | $0.50 / 1M tokens (Magistral Small); $2.00 / 1M tokens (Magistral Medium) | [2] |
| Output price | $1.50 / 1M tokens (Magistral Small); $5.00 / 1M tokens (Magistral Medium) | [2] |
| Licence | Magistral Small: Apache 2.0 (open weights). Magistral Medium: proprietary, API-only. · open weights | [1] |
| Released | 10 June 2025 | [3] |
| Parameters | Magistral Small: 24B (≈23.6B). Magistral Medium: undisclosed (built on Mistral Medium 3). | [1] |
| Knowledge cutoff | Not disclosed by Mistral. | [1] |

## Sources

1. [Magistral Small 2506 model card (Hugging Face)](https://huggingface.co/mistralai/Magistral-Small-2506)
2. [Mistral API pricing](https://mistral.ai/pricing/)
3. [Magistral announcement (Mistral AI)](https://mistral.ai/news/magistral/)

## Overview

Magistral 1.0 is Mistral AI's first family of reasoning models, announced on June 10, 2025. It ships in two sizes: Magistral Small, a 24-billion-parameter open-weight model released under the Apache 2.0 license (model ID magistral-small-2506), and Magistral Medium, a larger, more capable API-only model (magistral-medium-2506) available in preview on Le Chat and Mistral's La Plateforme API. Both are designed to 'think things through' — producing an explicit, verifiable reasoning trace rather than answering in a single step.

Magistral Small is built from Mistral Small 3.1 by adding reasoning through supervised fine-tuning on traces distilled from Magistral Medium, followed by reinforcement learning. Because it is 24B and Apache-2.0 licensed, it can be self-hosted and, in quantized form, run on a single high-memory consumer GPU. Magistral Medium is built on Mistral Medium 3 and trained with a from-scratch reinforcement-learning pipeline; Mistral highlights it for speed, claiming up to roughly 10x faster token throughput than many competitors when used through Le Chat's 'Flash Answers'.

At launch both 1.0 models are text-only and emphasize domain-specific, transparent, and multilingual reasoning, with strong support for languages including English, French, Spanish, German, Italian, Arabic, Russian, and Simplified Chinese. They are positioned for structured, multi-step work — calculations, programmatic logic, decision trees, and rule-based systems — where a traceable thought process matters. On math and science benchmarks the 1.0 models trail the strongest closed reasoning models of the period (e.g. Gemini 2.5 Pro, Claude Opus 4), but Magistral Small was notable as one of the few openly licensed reasoning models of its time.

## Benchmarks

- AIME 2024 (pass@1): 73.6% ([source](https://arxiv.org/html/2506.10910v1))
- AIME 2025 (pass@1): 64.9% ([source](https://arxiv.org/html/2506.10910v1))
- GPQA Diamond: 70.8% ([source](https://arxiv.org/html/2506.10910v1))
- LiveCodeBench v5: 59.4% ([source](https://arxiv.org/html/2506.10910v1))
- AIME 2024 (pass@1): 70.7% ([source](https://huggingface.co/mistralai/Magistral-Small-2506))
- AIME 2025 (pass@1): 62.8% ([source](https://arxiv.org/html/2506.10910v1))
- GPQA Diamond: 68.2% ([source](https://huggingface.co/mistralai/Magistral-Small-2506))
- LiveCodeBench v5: 55.8% ([source](https://huggingface.co/mistralai/Magistral-Small-2506))

## Pricing

- Input: $0.50 / 1M tokens (Magistral Small); $2.00 / 1M tokens (Magistral Medium) per 1M tokens
- Output: $1.50 / 1M tokens (Magistral Small); $5.00 / 1M tokens (Magistral Medium) per 1M tokens

Mistral La Plateforme API pricing. Reasoning models emit verbose chain-of-thought, so output-token usage (and cost) per request tends to be higher than non-reasoning models. Magistral Small weights are also free to self-host under Apache 2.0.

Source: https://mistral.ai/pricing/

## Strengths

- Magistral Small ships under Apache 2.0 with downloadable weights — fully self-hostable and commercially usable, rare for a reasoning model at launch
- Transparent, traceable chain-of-thought reasoning you can read and verify, rather than a single opaque answer
- Strong multilingual reasoning across 20+ languages including English, French, Spanish, German, Italian, Arabic, Russian, and Simplified Chinese
- Magistral Medium emphasizes throughput — Mistral claims up to ~10x faster token speed than many competitors via Le Chat Flash Answers
- Small is compact (24B), so it runs on a single high-memory GPU when quantized, keeping reasoning workloads on-prem
- Available across many endpoints (Mistral La Plateforme, Together AI, OpenRouter, NVIDIA NIM) plus self-deployment

## Best for

- Self-hosted, privacy-sensitive reasoning workloads using the open-weight Magistral Small
- Multi-step math, logic, and structured-calculation tasks that benefit from an explicit reasoning trace
- Multilingual reasoning and analysis across European, Arabic, Russian, and Chinese text
- Auditable enterprise decision support — decision trees, rule-based systems, and programmatic logic where the thought process must be inspectable
- Coding and software-reasoning tasks (evaluated on LiveCodeBench) within agentic or assistant pipelines
- Low-latency assistant experiences via Magistral Medium's high-throughput 'Flash Answers' on Le Chat

## How to access

- [Mistral La Plateforme](https://mistral.ai/pricing/) — model id `magistral-medium-2506`
- [Mistral La Plateforme](https://mistral.ai/pricing/) — model id `magistral-small-2506`
- [Hugging Face (self-host)](https://huggingface.co/mistralai/Magistral-Small-2506) — model id `mistralai/Magistral-Small-2506`
- [OpenRouter](https://openrouter.ai/mistralai/magistral-medium-2506) — model id `mistralai/magistral-medium-2506`
- [OpenRouter](https://openrouter.ai/mistralai/magistral-small-2506) — model id `mistralai/magistral-small-2506`
- [Together AI](https://www.together.ai/models/magistral-small-2506) — model id `magistral-small-2506`
- [NVIDIA NIM](https://build.nvidia.com/mistralai/magistral-small-2506) — model id `magistral-small-2506`

## Magistral versions

- [Magistral 1.2 (Small & Medium)](https://ai-tldr.dev/models/magistral-1-2/) (17 September 2025)
- [Magistral 1.1 (Small & Medium)](https://ai-tldr.dev/models/magistral-1-1/) (July 2025)
- Magistral 1.0 (Small & Medium) (10 June 2025) — this page

## FAQ

### What is the difference between Magistral Small and Magistral Medium?

Magistral Small is a 24B-parameter open-weight model released under Apache 2.0 — you can download and self-host it. Magistral Medium is a larger, more capable model available only through Mistral's API and Le Chat. Small is built from Mistral Small 3.1 (via fine-tuning on Medium's reasoning traces plus RL); Medium is built on Mistral Medium 3 with a pure reinforcement-learning pipeline and scores a few points higher on math, science, and coding benchmarks.

### Is Magistral 1.0 open source?

Magistral Small is open-weight under the Apache 2.0 license, which permits commercial use and self-hosting; the weights are on Hugging Face as mistralai/Magistral-Small-2506. Magistral Medium is proprietary and available only via API. Note that 'open weights' means the trained weights are released — the training data and full pipeline are not.

### What context window does Magistral 1.0 support?

Magistral Small supports up to a 128K-token context window, though Mistral recommends working within about 40K tokens because performance can degrade beyond that point.

### How much does Magistral cost to use via API?

On Mistral's La Plateforme, Magistral Small is $0.50 per million input tokens and $1.50 per million output tokens, while Magistral Medium is $2.00 input and $5.00 output per million tokens. Because reasoning models produce verbose chain-of-thought, output usage per request tends to be higher than with non-reasoning models. Magistral Small can also be self-hosted for free under Apache 2.0.

## Links

- [Magistral announcement (Mistral AI)](https://mistral.ai/news/magistral/)
- [Magistral Small 2506 model card (Hugging Face)](https://huggingface.co/mistralai/Magistral-Small-2506)
- [Magistral technical report (arXiv 2506.10910)](https://arxiv.org/html/2506.10910v1)
- [Mistral API pricing](https://mistral.ai/pricing/)
- [Magistral Medium 2506 on OpenRouter](https://openrouter.ai/mistralai/magistral-medium-2506)
- [Magistral Small 2506 on OpenRouter](https://openrouter.ai/mistralai/magistral-small-2506)
- [Mistral releases a pair of AI reasoning models (TechCrunch)](https://techcrunch.com/2025/06/10/mistral-releases-a-pair-of-ai-reasoning-models/)
- [Magistral — the first reasoning model by Mistral AI (Simon Willison)](https://simonwillison.net/2025/Jun/10/magistral/)

Source: AI/TLDR LLM registry, https://ai-tldr.dev/models/magistral-1-0/