兔子API 兔子API
中

provider

Mistral AI

Direct answerA first direct Mistral call uses the Mistral base URL, Authorization: Bearer, and an exact model ID. This site exposes only the adapted /v1/chat/completions route and requires this site's base URL and key. Batch, Guardrails, and Moderation are official Mistral capabilities, not evidence that this route automatically supports or enables them.

Updated · Reviewed

Separate the vendor API, this site's compatibility route, and open weights

Mistral AI is both a model developer and a hosted API platform, and it publishes some weights for self-hosting. These modes are not interchangeable: Mistral manages accounts, quota, and service terms for direct API use; this site uses its own credential to relay through configured channels; self-hosters own hardware, serving, safety, and weight-license compliance.

Bearer authentication and the first request

For direct access, create a workspace API key in Studio, send JSON to https://api.mistral.ai/v1/chat/completions, and set Authorization: Bearer <MISTRAL_API_KEY>. Start with one short prompt and a pinned model ID to isolate authentication, model-access, and parameter errors. This site exposes a route with the same /v1/chat/completions shape, but its base URL, key, and available model IDs come from this site's model marketplace.

Batch is an official asynchronous job API

Mistral Batch accepts inline requests or JSONL files whose records contain a unique custom_id and a body. File batches can contain up to one million requests, while inline batching is available below ten thousand requests. After creation, query the job ID through QUEUED, RUNNING, SUCCESS, FAILED, and related states, then read output after success. This page does not list Batch as a site route, so do not copy the official Batch URL onto this gateway.

RPS, TPM, and 429

Mistral displays Completion limits per model, including requests per second (RPS) and tokens per minute (TPM). Bursty short calls can exhaust RPS, while long contexts or sustained generation can exhaust TPM; either may result in 429. Production clients should read their organization's actual Limits, cap concurrency, use bounded backoff, and avoid infinite retries.

Guardrails and Moderation have different roles

Mistral recommends Custom Guardrails that can be declared for Chat Completions, Conversations, or Agent configurations. The separate Moderation API classifies raw text or conversations with a safety model and suits pipelines that need raw scores and custom thresholds. Applications still own false-positive and false-negative handling, appeals, logging, and human escalation.

Open weights do not share one unconditional license

Mistral says most open models use Apache 2.0, but some use a conditional modified MIT license under which companies above USD 20 million in monthly revenue need a commercial license or must use Mistral Studio. Use the exact model card's license and review weight, code, data, and hosted-service terms separately; a moving latest alias is not a reliable licensing or reproducibility record.

Reviewed provider catalog

Latest models by capability

These are provider-catalog models, not a promise of availability on this site. Confirm callable IDs, endpoints, and pricing in the model marketplace.

Text, reasoning, and code

Speech and audio

Embeddings, retrieval, and reranking

Check live availability in the model marketplace

Use cases

  • Multimodal agents
  • Coding and reasoning
  • OCR, speech, and safety

API protocols

  • /v1/chat/completions

FAQ

What should I check when my first Mistral API call returns 401?

For a direct call, verify that the request goes to https://api.mistral.ai/v1/chat/completions, uses Authorization: Bearer <MISTRAL_API_KEY>, and names an exact model available to the current workspace. Through this site, change both the base URL and key to this site's values; the two providers' keys are not interchangeable.

Does Mistral Batch return results synchronously, and is it supported by this site?

No. An official Batch job moves through states such as QUEUED and RUNNING; query it by job ID and read output after success. Requests can be inline or supplied as JSONL records with custom_id and body. Batch is absent from this page's api_paths, so the official /v1/batch/jobs URL is not a route on this site.

How do Mistral RPS and TPM differ, and why can I still receive 429?

RPS caps requests per second, while TPM caps a model's token throughput per minute. Exhausting either limit for the current organization and model can produce 429. Read the actual API Limits in the Admin Panel instead of hard-coding another account's plan values.

Are Mistral Guardrails and the Moderation API the same thing?

No. Mistral recommends Custom Guardrails declared for chat, Conversations, or Agent requests. The separate Moderation API returns safety-category classifications or scores for applications that own thresholding and workflow logic. Neither means this site automatically moderates every input and output.

Can every Mistral open-weight model be used commercially without conditions?

Do not assume so. Mistral says most open models use Apache 2.0, while some use a modified MIT license with a revenue threshold. Check the exact model card and license. Buying hosted API access also does not automatically grant rights to download, modify, or redistribute weights.

  • Mistral familyMistral currently features Medium 3.5 and Small 4 for general and agentic work, plus dedicated code, OCR, speech, embedding, and safety models with distinct licenses.
  • OpenAI-compatible API guideThis site's OpenAI-compatible routes use a Bearer token; Responses, Chat Completions, and media endpoints have distinct objects and streaming semantics and must match the target model.
  • RAG, embedding, and reranking model guideRAG combines chunking, embeddings, retrieval, reranking, grounded generation, and evaluation; completeness comes from citable evidence and no-answer handling, not a larger model list.
  • AI API quickstart: from API key to first requestAn AI API call needs a matching base URL, API key, model ID, and endpoint; start with minimal curl, then add error handling, cost controls, and key security.
  • Common AI API error codes and troubleshootingHandle AI API failures by parsing the HTTP status, provider code, error type, message, retry hints, and request ID together. Fix request, authentication, billing, and hard-quota causes first; retry only transient throttling, network failures, and selected 5xx errors with strict bounds.

Official sources

  1. Activate Mistral Studio and Make a First API Call Official
  2. Mistral Batch Processing Official
  3. Mistral Usage and Limits Official
  4. Mistral Moderation and Guardrailing Official
  5. Mistral Model Selection Guide Official
  6. Mistral Open Model Licensing Official
  7. Mistral API Reference Official