Last verified
EU FRONTIER LABFOUNDED 2023PARIS · FRANCEOPEN WEIGHTSLE CHAT MAKERS

Mistral API Pricing

The Paris lab behind Le Chat and the open-weight Mistral family. Founded in 2023 by Arthur Mensch with ex-DeepMind and Meta researchers, it ships frontier-grade models with Apache-2.0 open weights for most tiers and EU-native data residency — the leading European alternative to US labs.

Production models
14
14 GA
Founded
2023
3 years operating
Cheapest tier
Free/M
Leanstral input
Open weights
Apache 2.0
Small · Ministral · Devstral
Headquarters
Paris
EU · GDPR-native
Latest valuation
$14B
ASML-led, Sep 2025
§ 01 / LINEUP

The full roster.

Side-by-side →
FAST · LIGHTWEIGHT

Mistral Small 4

New standard for multimodal, reasoning-optimized open-weight models.

Input
$0.15/M
Output
$0.60/M
documented elsewhere — not on pricing page ctx · vision
FAST · LIGHTWEIGHT

Ministral 3 14B

API ID `ministral-3-14b-2512`.

Input
$0.20/M
Output
$0.20/M
128K ctx · vision
FAST · LIGHTWEIGHT

Ministral 3 8B

API ID `ministral-3-8b-2512`.

Input
$0.15/M
Output
$0.15/M
128K ctx · vision
BALANCED · MID-TIER

Mistral Medium 3.5

Higher list rate than Mistral Large 3 — positioned as state-of-the-art agentic/coding model.

Input
$2/M
Output
$8/M
documented elsewhere — not on pricing page ctx · vision
FAST · LIGHTWEIGHT

Ministral 3 3B

API ID `ministral-3-3b-2512`.

Input
$0.10/M
Output
$0.10/M
128K ctx · vision
FAST · LIGHTWEIGHT

Mistral Small 3.2

API ID `mistral-small-2506`.

Input
$0.10/M
Output
$0.30/M
documented elsewhere ctx · vision
BALANCED · MID-TIER

Mistral Medium 3.1

API ID `mistral-medium-2508`.

Input
$2/M
Output
$8/M
documented elsewhere ctx · vision
FRONTIER · REASONING

Mistral Large 3

Open-weight, general-purpose, flagship multimodal + multilingual.

Input
$0.50/M
Output
$2/M
documented elsewhere — not on pricing page ctx · vision
SPECIALIST

Devstral 2

Coding-agent specialist (advanced agentic coding model).

Input
$0.40/M
Output
$2/M
documented elsewhere — not on pricing page ctx · text-only
FAST · LIGHTWEIGHT

Devstral Small 2

API ID `labs-devstral-small-2512`.

Input
$0.10/M
Output
$0.30/M
documented elsewhere ctx · vision
SPECIALIST

Codestral

Premier coding model; supports 80+ programming languages.

Input
$0.30/M
Output
$0.90/M
documented elsewhere — not on pricing page ctx · text-only
FRONTIER · REASONING

Magistral Medium

Thinking/reasoning model — excels in domain-specific, transparent, multilingual reasoning.

Input
$2/M
Output
$5/M
documented elsewhere — not on pricing page ctx · text-only
FAST · LIGHTWEIGHT

Magistral Small

Premier tier reasoning model — lightweight Magistral sibling.

Input
$0.50/M
Output
$2/M
documented elsewhere ctx · text-only
FAST · LIGHTWEIGHT

Leanstral

API ID `leanstral-2603`.

Input
$0.00/M
Output
$0.00/M
documented elsewhere ctx · text-only
DEPRECATED · 2026-02-27

Mistral Large 2.1 LAUNCHED NOV 2024

API ID mistral-large-2411 (also referred to as Mistral Large 24.11).

Input
$2/M
Output
$6/M
128K ctx · text-only
DEPRECATED · 2026-02-27

Pixtral Large LAUNCHED NOV 2024

API ID pixtral-large-2411.

Input
$2/M
Output
$6/M
128K ctx · vision
LEGACY · STILL SUPPORTED

Mistral NeMo 12B LAUNCHED JUL 2024

API ID open-mistral-nemo.

Input
$0.15/M
Output
$0.15/M
128K ctx · text-only
RETIRED · NOV 2025

Codestral 2501 LAUNCHED JAN 2025

API ID codestral-2501.

Input
$0.30/M
Output
$0.90/M
256K ctx · text-only
RETIRED · JUN 2025

Codestral 2405 LAUNCHED MAY 2024

API ID codestral-2405.

Input
$0.20/M
Output
$0.60/M
32K ctx · text-only
RETIRED · DEC 2025

Pixtral 12B LAUNCHED SEP 2024

API ID pixtral-12b (pixtral-12b-2409).

Input
$0.15/M
Output
$0.15/M
128K ctx · vision
RETIRED · MAR 2025

Mixtral 8x22B LAUNCHED APR 2024

API ID open-mixtral-8x22b.

Input
$2/M
Output
$6/M
64K ctx · text-only
RETIRED · MAR 2025

Mixtral 8x7B LAUNCHED DEC 2023

API ID open-mixtral-8x7b.

Input
$0.70/M
Output
$0.70/M
32K ctx · text-only
RETIRED · DEC 2025

Ministral 8B LAUNCHED OCT 2024

API ID ministral-8b-2410 (Ministral 8B 24.10).

Input
$0.10/M
Output
$0.10/M
128K ctx · text-only
RETIRED · MAR 2025

Mistral 7B LAUNCHED SEP 2023

API ID open-mistral-7b.

Input
$0.25/M
Output
$0.25/M
32K ctx · text-only
RETIRED · DEC 2025

Ministral 3B LAUNCHED OCT 2024

API ID ministral-3b-2410 (Ministral 3B 24.10).

Input
$0.04/M
Output
$0.04/M
128K ctx · text-only
RETIRED · MAR 2025

Mistral Large 2 LAUNCHED JUL 2024

API ID mistral-large-2407 (Mistral Large 24.07).

Input
$2/M
Output
$6/M
128K ctx · text-only
RETIRED · JUN 2025

Mistral Large 1 LAUNCHED FEB 2024

API ID mistral-large-2402 (also called Mistral Large 24.02).

Input
$4/M
Output
$12/M
32K ctx · text-only
RETIRED · JUN 2025

Mistral Medium 1 LAUNCHED DEC 2023

API ID mistral-medium-2312.

Input
$3/M
Output
$8/M
32K ctx · text-only
RETIRED · JUN 2025

Mistral Small 1 LAUNCHED FEB 2024

API ID mistral-small-2402 (Mistral Small 24.02).

Input
$1/M
Output
$3/M
32K ctx · text-only
RETIRED · JUN 2025

Codestral Mamba 7B LAUNCHED JUL 2024

API ID open-codestral-mamba.

Input
$0.25/M
Output
$0.25/M
256K ctx · text-only
§ 02 / SHELF

All side-by-side.

Methodology →
Model Input /M Output /M Cached Context Max output Vision Tools Tier
Mistral Small 4 $0.15 $0.6 documented elsewhere — not on pricing page Light
Ministral 3 14B $0.2 $0.2 128K Light
Ministral 3 8B $0.15 $0.15 128K Light
Mistral Medium 3.5 $1.50 $7.50 documented elsewhere — not on pricing page Mid
Ministral 3 3B $0.1 $0.1 128K Light
Mistral Small 3.2 $0.1 $0.3 documented elsewhere Light
Mistral Medium 3.1 $1.50 $7.50 documented elsewhere Mid
Mistral Large 3 $0.5 $1.50 documented elsewhere — not on pricing page Frontier
Devstral 2 $0.4 $2.00 documented elsewhere — not on pricing page Active
Devstral Small 2 $0.1 $0.3 documented elsewhere Light
Codestral $0.3 $0.9 documented elsewhere — not on pricing page Active
Magistral Medium FLAGSHIP $2.00 $5.00 documented elsewhere — not on pricing page Frontier
Magistral Small $0.5 $1.50 documented elsewhere Light
Leanstral Free Free documented elsewhere Light
Mistral Large 2.1 DEPRECATED $2.00 $6.00 128K Deprecated
Pixtral Large DEPRECATED $2.00 $6.00 128K Deprecated
Mistral NeMo 12B LEGACY $0.15 $0.15 128K Legacy
Codestral 2501 RETIRED Retired Nov 30, 2025 → codestral Retired
Codestral 2405 RETIRED Retired Jun 16, 2025 → codestral Retired
Pixtral 12B RETIRED Retired Dec 31, 2025 → ministral-3-14b Retired
Mixtral 8x22B RETIRED Retired Mar 30, 2025 → mistral-small-4 Retired
Mixtral 8x7B RETIRED Retired Mar 30, 2025 → mistral-small-4 Retired
Ministral 8B RETIRED Retired Dec 31, 2025 → ministral-3-8b Retired
Mistral 7B RETIRED Retired Mar 30, 2025 → ministral-3-8b Retired
Ministral 3B RETIRED Retired Dec 31, 2025 → ministral-3-3b Retired
Mistral Large 2 RETIRED Retired Mar 30, 2025 → mistral-large-3 Retired
Mistral Large 1 RETIRED Retired Jun 16, 2025 → mistral-large-3 Retired
Mistral Medium 1 RETIRED Retired Jun 16, 2025 → mistral-medium-3-5 Retired
Mistral Small 1 RETIRED Retired Jun 16, 2025 → mistral-small-4 Retired
Codestral Mamba 7B RETIRED Retired Jun 6, 2025 → codestral Retired

The last 12 months at Mistral.

Major releases, funding, and EU-infrastructure moves. Sourced from official Mistral announcements + press coverage, verified at publication.

APR 29 · 2026
Mistral Medium 3.5 released — agentic/coding multimodal mid-tier at $1.50/$7.50, powers Le Chat Work mode
RELEASE
APR · 2026
Le Chat Enterprise + Work mode launched — EU data residency, remote coding agents, memory
FEATURE
MAR 30 · 2026
$830M debt financing raised to build the Essonne (Paris-region) Nvidia data-center cluster
CORPORATE
MAR 16 · 2026
Mistral Small 4 launched — cheap open multimodal tier at $0.15/$0.60, 256K context
RELEASE
DEC 09 · 2025
Devstral 2 released — frontier software-engineering code agent, 256K context
RELEASE
DEC 02 · 2025
Mistral Large 3 released — open-weight flagship, multimodal, $0.50/$1.50 per M
RELEASE
SEP 18 · 2025
Magistral Medium 1.2 shipped — transparent, multilingual reasoning model at $2/$5
RELEASE
SEP 09 · 2025
Series C: €1.7B raised, ASML-led — valuation more than doubles to ~$14B
CORPORATE
AUG · 2025
Codestral 25.08 card updated — FIM coding model across 80+ languages
RELEASE
JUN · 2025
Mistral Compute unveiled at VivaTech — sovereign EU AI infrastructure with Nvidia (18K Blackwell GPUs)
CORPORATE
§ 04 / ACCESS

Where to get it.

Methodology →
§ 04 / BEST FOR

Which Mistral AI for what.

More scenarios →
If you need an open-weight multimodal flagship
Mistral Large 3
Profile →
If you're building agentic or coding workflows…
Mistral Medium 3.5
Profile →
If you want transparent multilingual reasoning
Magistral Medium
Profile →
If you need code completion (FIM) in an IDE…
Codestral
Profile →
If you run software-engineering agents
Devstral 2
Profile →
If you want cheap high-volume multimodal
Mistral Small 4
Profile →
If you require EU data residency for compliance…
Le Chat Enterprise
Mistral docs →
If you write Lean 4 formal proofs for free…
Leanstral
Profile →
§ 06 / BACKGROUND

The company behind it.

mistral.ai →

Mistral AI was founded in April 2023 in Paris by Arthur Mensch (CEO, ex-Google DeepMind), Guillaume Lample (Chief Scientist) and Timothée Lacroix (CTO) — the latter two from Meta's Llama team. The pitch: build frontier-grade models in Europe with open weights and EU data sovereignty as core principles rather than afterthoughts.

The company is a private French société par actions simplifiée (SAS). Funding has been rapid: a €105M seed in mid-2023, a ~$2B-valuation Series A in late 2023, a ~$6B round in 2024, and a €1.7B Series C in September 2025 led by chip-equipment maker ASML that more than doubled its valuation to roughly $14B. Investors include Andreessen Horowitz, Nvidia, Lightspeed, General Catalyst, DST Global, Index Ventures and Bpifrance.

Mistral's technical signature is its open-weight release strategy: the Mixtral mixture-of-experts models, the Small and Ministral families, and Devstral Small ship under Apache 2.0 licenses on Hugging Face — fully self-hostable. Premier models like Mistral Medium 3.5 and Magistral Medium remain API-only. The lab also builds the Le Chat consumer assistant and the Codestral / Devstral coding line.

Beyond models, Mistral is investing in European AI sovereignty. Mistral Compute, unveiled at VivaTech in June 2025 with Nvidia, provides sovereign GPU infrastructure so EU companies and governments can own their AI stack without depending on US or Chinese cloud providers; the lab raised $830M in debt in 2026 to build a Paris-region data-center cluster.

Versus US labs: Mistral's frontier benchmarks trail OpenAI and Anthropic, and its lineup is narrower. But it is materially cheaper, ships open weights, and is the clearest EU-native, GDPR-first option — a meaningful differentiator for European procurement and regulated industries.

§ 07 / COMPETITORS

Other frontier labs.

All providers →
PRIMARY RIVAL

OpenAI

Wider lineup + ecosystem (GPT-5 family, ChatGPT, Azure). Stronger frontier benchmarks. But closed weights, US data residency, and pricier output.

GPT-5 family
FRONTIER LEADER

Anthropic

Leads coding + reasoning benchmarks with Claude. Zero-retention defaults. But no open weights, higher prices, and no EU-native data residency by default.

Claude lineup

Frequently asked.

Practical questions about Mistral AI, open weights, and EU data residency.

Q · 01 Which Mistral model should I start with? +
For most teams: Mistral Medium 3.5 ($1.5/$7.5) for agentic and coding work, or Mistral Large 3 ($0.5/$1.5) if you want the open-weight flagship. Drop to Mistral Small 4 ($0.15/$0.60) for cheap high-volume multimodal. See the use case picker above.
Q · 02 Is Mistral a good choice for EU data residency? +
Yes — this is Mistral's core differentiator. The company is Paris-based and GDPR-native. Le Chat Enterprise offers EU data residency, and Mistral Compute provides sovereign EU GPU infrastructure built with Nvidia. For regulated European industries, it avoids the US data-transfer questions that come with American labs.
Q · 03 Are Mistral models open-weight? +
Many are. The Small, Ministral, Devstral Small, and Mixtral families ship under Apache 2.0 on Hugging Face — fully self-hostable with no per-token fee. Premier models like Mistral Medium 3.5 and Magistral Medium are API-only. Check each model card before assuming a license.
Q · 04 What's the data retention policy? +
API requests on La Plateforme are not used to train models by default. Le Chat Enterprise adds stricter data-residency and retention controls for business customers. Free consumer tiers may retain data for product improvement unless you opt out.
Q · 05 Is there a free option? +
Yes — Leanstral, a Lean 4 formal-proof code agent, is a $0/M free API row on La Plateforme. The cheapest paid general model is Ministral 3 3B at $0.10/M (symmetric input/output). Open weights also let you self-host any Apache-2.0 model for compute cost only.
Q · 06 Can I get a discount? +
Mistral's Batch API gives −50% on async workloads. There is no separate cached-input SKU on the pricing page, so the calculators here treat cached input at the base rate. Priority-tier (low-latency) APIs run about +75% over list. Self-hosting open weights removes per-token cost entirely.
Q · 07 How does Mistral compare to OpenAI, Anthropic, and DeepSeek? +
Mistral trails OpenAI and Anthropic on top frontier benchmarks and has a narrower lineup. But it is cheaper, ships open weights, and is the clearest EU-native choice. Against DeepSeek, prices are comparable but Mistral wins on data governance for European teams.