"Frontier intelligence, customized to you."
The Mistral API is the developer and enterprise interface for Mistral models.
Through Mistral AI Studio, companies and developers can use models via API, test prompts, build agents, implement RAG workflows, use fine-tuning, manage workspaces, and bill API usage. Mistral offers both open-weight and commercial/premier models.
Mistral API
LLM - build, customize, and deploy AI, your way
Location: France ⓘ Mistral AI, 15 rue des Halles, 75001 Paris, France. Mistral is registered in Paris under number 952 418 325.
Self-Deployment / Open-Weight Models Selected models can be operated independently or via cloud/enterprise deployments; the range of features depends on the respective model.
Enterprise Private Deployment Customized private deployment for organizations with increased control, security, and scalability requirements.
Target audience
The Mistral API is aimed at developers, start-ups, software teams, agencies, AI product teams, SMEs, corporations, public institutions, and regulated organizations that want to integrate generative AI into their own products or internal systems. Mistral is particularly relevant for European companies that value data location, AVV/DPA, flexible deployment models, and open-weight options. Typical roles include developers, CTOs, data/AI teams, product managers, compliance officers, and IT architects.
Outstanding features
Outstanding features include the combination of hosted API, workspaces, API keys, spend limits, pay-as-you-go billing, fine-tuning/customization options, agents, function calling, structured outputs, RAG workflows, OCR, embeddings, moderation, and coding models. Mistral also offers several deployment paths: Managed Mistral Cloud, cloud provider integrations, Mistral Compute, self-deployment, VPC, edge, and on-premises.
Most important application areas
The API is suitable for chatbots, internal knowledge assistants, RAG systems, document analysis, OCR workflows, code assistance, software agents, automated text generation, translation, classification, moderation, semantic search, data extraction, agents with tools, customer service automation, and AI features in SaaS products. With models such as Magistral and Devstral, Mistral also covers reasoning and software development scenarios; with Mistral Large 3, Medium 3.1, and Small 4, multimodal and high-performance generalist models are available.
Usage & notes
To use it, a workspace is created in Mistral AI Studio, an API key is generated, and then work is carried out via API, SDKs, or Playground. For production systems, the Experiment plan should not be used; instead, the Scale plan or an Enterprise contract should be chosen. For GDPR-critical scenarios, organizations should at least review the DPA, region, subprocessors, data retention, training status, ZDR, logging, access controls, spend limits, and, if necessary, self-deployment. Sensitive personal data should only be processed if the legal basis, AVV, TOMs, deletion concept, and data flows are properly documented.
| Target audience | Assessment |
|---|---|
| Developers / software teams | Very suitable – for chat, coding, agents, RAG, structured outputs, embeddings, OCR, and multimodal AI applications. |
| EU companies / GDPR-oriented teams | Very suitable – especially because of EU hosting as standard, DPA, API no-training, and a European provider profile. |
| SaaS providers / product teams | Very suitable – if AI features are to be integrated quickly into their own products via API. |
| SMEs with technical resources | Suitable – for customer service, document analysis, internal search, automation, and knowledge management. |
| Large enterprises | Very suitable – because of enterprise options, private deployments, admin/team features, and self-deployment paths. |
| Private individuals without a technical background | Rather not suitable for the API – Le Chat is more appropriate for them; the API requires technical integration. |
Calculate tokens and costs with the KIFOX Tokenizer
Mistral Medium 3.5
Particularly suitable for demanding coding, agent, and productivity workflows, such as multi-step tasks with tool use, structured output, reasoning, code generation, and multimodal processing with a large context window
Mistral Large 3
Frontier generalist, multimodality, complex enterprise workflows, long contexts, agents, tool use, demanding text/image tasks
Mistral Medium 3.1
strong all-rounder, multimodal business apps, good price-performance ratio, chatbots, RAG, agents, structured outputs
Mistral Medium 3
older but still supported Medium generation, multimodal applications, general enterprise workflows
Mistral Small 4
cost-effective productive scaling, hybrid tasks, instruct + reasoning + coding, long contexts, high request volumes
Mistral Small 3.2
efficient standard tasks, fast chatbots, classification, summaries, simple RAG use cases
Ministral 3 14B
compact multimodal workloads, self-hosting, edge/VPC scenarios, good balance of quality and cost
Ministral 3 8B
efficient local/private cloud use, simple assistants, classification, cost-sensitive applications
Ministral 3 3B
very small deployments, edge, embedded AI, routing, simple classifications, low latency
Magistral Medium 1.2
reasoning, complex inferences, multi-step analyses, planning tasks, demanding problem-solving
Magistral Small 1.2
more affordable reasoning, mathematical-logical tasks, structured problem-solving, self-hosting-adjacent scenarios
Devstral 2
software engineering agents, codebase analysis, multi-file edits, tool use, developer automation
Codestral
code completion, IDE integration, fill-in-the-middle, developer productivity, fast code suggestions
Leanstral
Lean 4 proofs, formal verification, mathematical proof engineering workflows
Voxtral Small
audio input, speech understanding, voice agents, audio-based assistance, multimodal audio/text tasks
Voxtral Mini Transcribe
transcription, speech-to-text, audio logs, cost-effective speech processing
Voxtral Mini Transcribe 2
newer transcription, speech-to-text, audio pipelines, higher efficiency
Voxtral Mini Transcribe Realtime
live transcription, streaming audio, real-time subtitles, voice interfaces
Mistral Nemo 12B
multilingual open-weight applications, local deployments, cost-efficient text tasks, older but still supported workloads
Hosting & Data
1) On-prem / local hosting
Meaning: The company operates the solution on its own hardware or within its own infrastructure. In the strictest sense, not only the application runs locally, but ideally the model as well.
2) Private cloud / data center
Meaning: The solution runs in a dedicated or more clearly separated cloud environment, often with a hosting provider or hyperscaler, but in a German data center or in a particularly controlled environment.
3) EU SaaS / managed
Meaning: The provider operates the solution itself as a service. The company uses the tool as a ready-made cloud service, ideally with EU data residency.
4) Hybrid
Meaning: One part of the processing remains internal / local / in a private cloud, while another part runs in an external cloud or EU SaaS.
5) AVV / DPA
Meaning: This is the data processing agreement or Data Processing Addendum. It governs that the provider processes personal data on behalf of the customer and is bound by the customer's instructions.
6) No training
Meaning: The provider does not use your prompts, uploads, attachments, chat histories, or outputs for training or improving the general model — ideally excluded by contract.
7) Open-source / transparency path
Meaning: There is a path toward greater technical transparency and sovereignty, for example through:
- open models
- documented components
- self-hostable parts
- traceable architecture
- export / switching options
| On-prem / local hosting | ✅ |
| Private cloud / data center | ⚠️ |
| EU SaaS / Managed | ⚠️ |
| Hybrid | ⚠️ |
| DPA / AVV | ✅ |
| No training on customer data | ⚠️ |
| Open source / transparency path | ✅ |
On-premises / local hosting: supported
The documentation explicitly describes “self-deployment”: Mistral models can be run on your own infrastructure. It also describes the local deployment of OpenWeight models on your own hardware.
Private Cloud / Data Center: Partially
Mistral Compute describes a private or isolated infrastructure with dedicated GPU clusters, bare-metal servers, managed clusters, and “Private AI Studio” in EU data centers. Whether this option is generally widely available for the Mistral API and exactly how it is structured contractually is only partially documented on the pages found.
EU SaaS / Managed: Partially
An EU-hosted managed offering is documented for Mistral Compute with “European-hosted AI cloud” and “EU Tier 3+ data centers.” However, for standard Mistral API/Studio SaaS usage, no generally binding EU/EEA data residency status for all data was found on the website.
Hybrid: Partially
The website describes flexible transitions between bare metal, managed clusters, Private AI Studio, and APIs, as well as local model deployment on the customer’s own infrastructure. This technically supports a hybrid approach, but a standard product explicitly described as “hybrid” for the Mistral API could not be clearly identified.
T&C / DPA: Covered
A Data Processing Addendum has been published. It describes Mistral as a data processor, bound by documented customer instructions and obligated to comply with data protection obligations, provide support, undergo audits, implement security measures, and adhere to subprocessor regulations.
No training: partially
The privacy documentation states that API data is not used for model training. At the same time, the DPA lists scenarios in which training is permitted under the Privacy Policy, provided that no opt-out applies or a product is not already excluded by default. Thus, the training-free path is well documented but not uniformly consistent across all products and pricing plans.
Open Source / Transparency Path: Covered
Mistral documents open models, self-deployment, and open-weight licenses. The documentation mentions local deployment via vLLM, TensorRT-LLM, and TGI. This establishes a clear path to transparency and data sovereignty.
Data Processing
The website outlines several processing models: standard use via API/Studio, European-hosted compute/private cloud variants, and self-deployment of open models on users’ own infrastructure. For business use, Mistral provides a Data Processing Agreement (DPA) and describes the use of subprocessors, including notification of changes. Standard Contractual Clauses (SCCs) are specified for international data transfers. It is stated that API data is not used for model training; additionally, opt-out options and data protection rights are available in the admin area.
Conclusion
For an EU/EEA-focused directory, Mistral API should not be evaluated as a standard SaaS solution that is unequivocally GDPR-compliant across the board, but it can be made well-suited for GDPR-compliant use under certain conditions. The strongest positive factor is the documented self-deployment/on-premises path for the open models, as well as the European-hosted compute offering. Less clear is the publicly available information regarding general EU data residency and the specific standard server locations for the regular API/Studio SaaS. Therefore, the overall rating is “conditional.”
Sources
- https://legal.mistral.ai/terms/privacy-policy
- https://legal.mistral.ai/terms/data-processing-addendum
- https://docs.mistral.ai/admin/security-access/privacy
- https://docs.mistral.ai/models/deployment
- https://docs.mistral.ai/models/deployment/local-deployment
- https://mistral.ai/products/compute/
- https://help.mistral.ai/en/articles/347629-where-do-you-store-my-data-or-my-organization-s-data
- https://help.mistral.ai/en/articles/347638-do-you-have-soc-2-or-iso-27001-certification
- https://help.mistral.ai/en/articles/347393-under-which-license-are-mistral-s-open-models-available
| On-prem / local hosting | ✅ |
| Private cloud / data center | ⚠️ |
| EU SaaS / Managed | ⚠️ |
| Hybrid | ⚠️ |
| DPA / AVV | ✅ |
| No training on customer data | ⚠️ |
| Open source / transparency path | ✅ |
On-premises / local hosting: supported
The documentation explicitly describes “self-deployment”: Mistral models can be run on your own infrastructure. It also describes the local deployment of OpenWeight models on your own hardware.
Private Cloud / Data Center: Partially
Mistral Compute describes a private or isolated infrastructure with dedicated GPU clusters, bare-metal servers, managed clusters, and “Private AI Studio” in EU data centers. Whether this option is generally widely available for the Mistral API and exactly how it is structured contractually is only partially documented on the pages found.
EU SaaS / Managed: Partially
An EU-hosted managed offering is documented for Mistral Compute with “European-hosted AI cloud” and “EU Tier 3+ data centers.” However, for standard Mistral API/Studio SaaS usage, no generally binding EU/EEA data residency status for all data was found on the website.
Hybrid: Partially
The website describes flexible transitions between bare metal, managed clusters, Private AI Studio, and APIs, as well as local model deployment on the customer’s own infrastructure. This technically supports a hybrid approach, but a standard product explicitly described as “hybrid” for the Mistral API could not be clearly identified.
T&C / DPA: Covered
A Data Processing Addendum has been published. It describes Mistral as a data processor, bound by documented customer instructions and obligated to comply with data protection obligations, provide support, undergo audits, implement security measures, and adhere to subprocessor regulations.
No training: partially
The privacy documentation states that API data is not used for model training. At the same time, the DPA lists scenarios in which training is permitted under the Privacy Policy, provided that no opt-out applies or a product is not already excluded by default. Thus, the training-free path is well documented but not uniformly consistent across all products and pricing plans.
Open Source / Transparency Path: Covered
Mistral documents open models, self-deployment, and open-weight licenses. The documentation mentions local deployment via vLLM, TensorRT-LLM, and TGI. This establishes a clear path to transparency and data sovereignty.
Data Processing
The website outlines several processing models: standard use via API/Studio, European-hosted compute/private cloud variants, and self-deployment of open models on users’ own infrastructure. For business use, Mistral provides a Data Processing Agreement (DPA) and describes the use of subprocessors, including notification of changes. Standard Contractual Clauses (SCCs) are specified for international data transfers. It is stated that API data is not used for model training; additionally, opt-out options and data protection rights are available in the admin area.
Conclusion
For an EU/EEA-focused directory, Mistral API should not be evaluated as a standard SaaS solution that is unequivocally GDPR-compliant across the board, but it can be made well-suited for GDPR-compliant use under certain conditions. The strongest positive factor is the documented self-deployment/on-premises path for the open models, as well as the European-hosted compute offering. Less clear is the publicly available information regarding general EU data residency and the specific standard server locations for the regular API/Studio SaaS. Therefore, the overall rating is “conditional.”
Sources
- https://legal.mistral.ai/terms/privacy-policy
- https://legal.mistral.ai/terms/data-processing-addendum
- https://docs.mistral.ai/admin/security-access/privacy
- https://docs.mistral.ai/models/deployment
- https://docs.mistral.ai/models/deployment/local-deployment
- https://mistral.ai/products/compute/
- https://help.mistral.ai/en/articles/347629-where-do-you-store-my-data-or-my-organization-s-data
- https://help.mistral.ai/en/articles/347638-do-you-have-soc-2-or-iso-27001-certification
- https://help.mistral.ai/en/articles/347393-under-which-license-are-mistral-s-open-models-available
Strengths & weaknesses at a glance
| Strengths | Weaknesses |
|---|---|
| • European provider based in France. | • The pricing page is publicly somewhat difficult to read by machine; specific API prices are often more reliably visible via individual model cards. |
| • EU hosting for data by default according to the Help Center. | • The free API experiment plan is intended only for evaluation/prototyping. |
| • DPA/AVV publicly available. | • Zero Data Retention for Mistral AI Studio is available only upon request and after review, not automatically. |
| • Scale plan data is not used for training according to the Help Center. | • Depending on the feature, data may be processed temporarily outside the EU; subprocessors must be reviewed. |
| • Open-weight and commercial models available. | • Not all models are open-weight; some are Premier/commercial. |
| • Flexible deployment: Mistral Cloud, cloud provider, VPC, on-premises, edge, self-deployment. | |
| • Broad model range: generalists, reasoning, code, multimodal, audio, OCR, moderation, embeddings. |
Reviews
0 reviews in total
There are no confirmed reviews for this tool yet.
Submit review
Your review will only become visible after email confirmation. This protects the portal against abuse.
Report review
Please select the reason why this review should be checked.
GDPR-compliant usage possible?
For users in the EU/EEA, GDPR-compliant use of the website is generally possible, but this is not consistently and clearly documented in the standard SaaS version. Positive aspects include a published Data Processing Agreement (DPA), explicitly described GDPR roles and data subject rights, documented sub-processor arrangements, and a clear self-deployment/on-premises path for open models. At the same time, for general API/Studio use on the website, there is no clearly documented EU/EEA server location established as the standard for all data processing; furthermore, the website mentions international data transfers using SCCs. Therefore, usage is considered “conditional” rather than clearly and fully substantiated.
Positive
Mistral publishes a privacy policy and a Data Processing Addendum (DPA). In the DPA, Mistral AI is described as a data processor for business use and commits to processing personal data only in accordance with the customer’s documented instructions. Support is provided for data subject rights, DPIA, audits, and data breach notifications. Regarding the API, the documentation states that data sent via the API is not used for model training. Additionally, there is self-deployment for models on the customer’s own infrastructure, as well as a European-hosted compute offering with EU data centers.
Negative
The website does not clearly specify a concrete, binding EU/EEA server location for general Mistral API/Studio SaaS use, nor does it specify a general EU data residency for all customer data. Instead, Mistral also describes international data transfers and refers to SCCs. A specific, publicly accessible list of subprocessors with individual locations was not directly found on the pages reviewed. For some training and model improvement uses, there are opt-out mechanisms or standard exceptions depending on the product and pricing plan, which require additional review without proper configuration.
Server Location
Not specifically stated on the website for general API/Studio data processing. The following was found: Mistral Compute is described as a European-hosted AI cloud with “EU Tier 3+ data centers”; a location in Sweden is also mentioned. Mistral cites SCCs for data transfers outside the EU.