Mistral API

Provides access to a tiered model family (Mistral Large, Medium, Small) via unified API endpoint, allowing developers to select models based on latency/cost tradeoffs without changing integration code. Models are served through Mistral's inference infrastructure with support for both streaming and batch completion modes, enabling real-time chat applications and asynchronous processing pipelines.

Implements OpenAI-compatible function calling where models receive a JSON schema describing available tools and can request tool invocation by returning structured function calls. Mistral's implementation uses a native function-calling layer that parses model outputs into structured tool requests, supporting both single and parallel function calls within a single generation step.

Provides token counting endpoints that allow developers to estimate token usage and costs before making API calls. This enables budget-aware applications that can make routing decisions based on estimated costs, implement cost limits, or optimize prompts to reduce token consumption.

Provides API key management through the console with granular rate limiting controls, allowing developers to create multiple keys with different rate limits, monitor usage, and implement quota-based access control. Rate limits are enforced per-key and per-model, enabling multi-tenant applications to allocate quotas to different users or services.

Constrains model outputs to valid JSON matching a provided schema, using guided generation techniques to ensure the model produces only valid, schema-compliant JSON without post-processing. The implementation uses token-level constraints during decoding to prevent invalid JSON syntax and enforce field requirements, eliminating the need for output parsing and validation.

Pixtral model enables multimodal understanding of images and text in a single request, supporting image analysis, OCR, visual question-answering, and image-to-text tasks. Images are encoded and processed alongside text prompts through the same unified API, allowing developers to build vision applications without separate image processing pipelines.

Codestral is a specialized code generation model optimized for programming tasks, supporting code completion, generation from natural language, code review, and debugging. It handles multiple programming languages and integrates with IDE plugins for inline code completion, providing context-aware suggestions based on file content and cursor position.

Enables training custom versions of Mistral models on proprietary datasets to adapt model behavior, domain knowledge, or output style. Fine-tuning uses supervised learning on labeled examples, updating model weights to specialize for specific tasks or domains. Mistral provides managed fine-tuning infrastructure, handling data validation, training, and model deployment.

Supports real-time token streaming via Server-Sent Events (SSE), allowing clients to receive model outputs incrementally as tokens are generated rather than waiting for full completion. This enables responsive chat interfaces, live transcription-like experiences, and reduced perceived latency in user-facing applications.

Provides asynchronous batch processing where developers submit multiple requests in a single batch job, receive a job ID, and poll for results. Batch processing is optimized for throughput rather than latency, offering lower per-token costs in exchange for delayed results (typically processed within hours).

Mistral operates infrastructure in the EU with explicit data residency guarantees, ensuring that user data and model inference remain within EU borders. This addresses GDPR compliance requirements and data sovereignty concerns for European organizations, with transparent data handling policies and no data sharing with third parties.

Manages multi-turn conversations by accepting a messages array containing full conversation history (system, user, assistant messages), allowing models to maintain context across multiple exchanges. The API handles message ordering, role-based formatting, and context window management, enabling stateless conversation APIs where clients maintain history.