Mistral

Run batches on Mistral's Batch API — a JSONL upload followed by a job that carries the model.

Mistral’s Batch API uploads a JSONL file and then creates a job that references it. The defining difference from OpenAI: the model and endpoint live on the job, not on each request line — so Batchwork strips model out of every line and sets it once on the job.

Example

import { batch } from "batchwork";
import { mistral } from "@ai-sdk/mistral";

const job = await batch({
  model: mistral("mistral-large-latest"),
  requests: [
    { customId: "fr", prompt: "Capital of France? One word." },
    { customId: "jp", prompt: "Capital of Japan? One word." },
  ],
});

How it works

1. Build — each request becomes a { custom_id, body } JSONL line, with model removed.
2. Upload — the JSONL is uploaded via the Files API (purpose=batch).
3. Create — POST /v1/batch/jobs sets the model, endpoint, and input_files on the job.
4. Poll — Mistral’s job status (QUEUED, RUNNING, SUCCESS, FAILED, TIMEOUT_EXCEEDED, …) maps onto the normalized status.
5. Read — the OpenAI-shaped output_file and error_file are streamed and parsed into BatchResults.

Batchwork validates Mistral job and file ids before using them in API paths.

Embeddings

Mistral’s mistral-embed model batches through the same job flow — the model is set on the job, and each request embeds one value. Submit with batch.embeddings():

import { batch } from "batchwork";
import { mistral } from "@ai-sdk/mistral";

const job = await batch.embeddings({
  model: mistral.embeddingModel("mistral-embed"),
  requests: [{ customId: "a", value: "The quick brown fox." }],
});

The vector lands on result.embedding. Mistral has no image model, so batch.images() throws UnsupportedProviderError.

Notes

Mistral has no native batch webhook, so it is poll-only. Use the server layer for unified completion events.

Credentials

Set MISTRAL_API_KEY, or pass apiKey / baseURL / headers to batch(). Install the peer dependency:

bun add @ai-sdk/mistral