Everyday workflows with Mistral (and what they cost)

This lesson gives you five workflows to run this week. Each names the model that fits, what it costs on Le Chat Free (zero, subject to limits) and what it costs on the API at catalogue prices. The API figures are estimates from token counts, so treat them as order-of-magnitude, not invoices. All of them are small; the point is to notice which one you would actually repeat every day.

1. Writing: draft, tighten, adapt

Paste your notes and ask for a first draft with a named audience and length. Then ask for a tighter version and a short social adaptation, in the same chat so the model keeps the facts. On Le Chat Free this is three prompts and costs nothing. On the API with Mistral Small 4, a 1,500-word draft is about 3K tokens in and 2K out, well under a cent. Even on Mistral Large 3 it is roughly half a cent.

2. Research: read the document so you do not have to

Upload a report in Le Chat, ask for the structured summary from Lesson 5, then ask follow-up questions against the same document. Web search adds current facts but counts against limits faster. On the API, a 50K-token document plus a 3K-token answer on Mistral Large 3 costs about three cents. Do not upload the same document every day; keep the chat and ask again.

3. Coding: Codestral in the editor, Devstral for the agent

For everyday coding, paste one function and one error, not the file. Le Chat Free handles this. In an editor plugin, Codestral 2508 is built for autocomplete and fill-in-the-middle; at $0.30 in / $0.90 out per 1M tokens a busy day of completions is typically a few cents. For an agent that edits many files and runs tools, Devstral 2 is the intended model, and because agents loop, expect ten to fifty times the tokens of a single question. That is the workflow that most often surprises people on the bill.

4. Data: classify, extract, normalise

Give the model ten rows and the output schema, check the result, then run the rest through the API. This is where Mistral's small models shine. Classifying 200,000 short records at about 100 tokens each is roughly 20M input tokens and 2M output tokens; on Ministral 3 8B ($0.15 / $0.15) that is around $3.30, and on Mistral Nemo it is under 50 cents. Le Chat is for the first ten rows only; do not paste spreadsheets into a chat window.

5. Planning: turn a goal into a schedule

Describe the goal, the deadline, the constraints and the people, and ask for a week-by-week plan as a table, then ask it to flag the riskiest week. This is a short-output task where any Mistral model does fine. Free on Le Chat; fractions of a cent on the API. If this is your main use, you will never need to pay.