← Back to API models

Mistral Large 3 & the Mistral lineup

by Mistral AI · European flagship family

Europe's frontier lab. Large 3 leads a lineup of cheap, focused models — Codestral for code, Pixtral for vision, Small for volume.

Input $2 / 1M
Output $6 / 1M
Context 256K tokens
Models 4 on our ledger
La Plateforme ↗ Updated June 10, 2026
§ API pricing

The full Mistral lineup.

Large 3
$2 / $6/1M tokens
Flagship · 256K context
  • Mistral's frontier general model
  • Same sticker as Grok 4.20
  • Strong multilingual, EU data residency
Codestral
$0.30 / $0.90/1M tokens
Code · 256K context
  • Purpose-built for code generation
  • Fill-in-the-middle completion support
  • Popular as a self-hostable IDE backend
Small 3.1
$0.20 / $0.60/1M tokens
Mid-tier · 128K context
  • General-purpose volume workhorse
  • Open-weights sibling available
  • Competes with GPT-5.4 mini on price
Pixtral 12B
$0.15 / $0.15/1M tokens
Vision · 128K context
  • Budget multimodal (image + text)
  • Flat rate — same price in and out
  • Cheapest vision input on our ledger

Why Mistral exists on this ledger

Mistral AI is the European counterweight to the US labs: a Paris-based frontier shop with an open-weights heritage and a paid API (La Plateforme) that undercuts the majors. It has no consumer subscription in our main comparison — Le Chat exists, but the API is the product that matters here. For teams with EU data-residency requirements, Mistral is frequently the shortlist by default.

Mistral Large 3, the flagship, sits at $2 input and $6 output per 1M tokens with a 256K window — the same sticker as Grok 4.20 and well under the Claude/GPT flagship tiers. The honest positioning: it's not benchmark-competitive with Fable 5 or GPT-5.5 at the very top, but for most production work the gap is smaller than the 2–8× price difference.

The specialist models

Codestral ($0.30/$0.90) is the one with a real fan base: a code-specialist with fill-in-the-middle support that slots into editors as a completion backend at a fraction of GPT or Claude rates. Small 3.1 ($0.20/$0.60) is the volume workhorse for classification, extraction, and routine generation — same price class as GPT-5.4 mini. Pixtral 12B ($0.15 flat both ways) is the budget vision option: when you need to look at thousands of images cheaply, its input rate is the lowest multimodal price on our table.

Typical use cases

  • EU data-residency and sovereignty-sensitive deployments
  • IDE code completion at volume via Codestral
  • High-volume classification and extraction on Small 3.1
  • Budget image understanding pipelines on Pixtral 12B
  • Multilingual European-language products on Large 3

Rival comparison

ModelInput / 1MOutput / 1MContext
Mistral Large 3$2$6256K
Grok 4.20$2$6256K
Gemini 3.1 Pro$2$121M
Mistral Small 3.1$0.20$0.60128K
GPT-5.4 mini$0.25$2272K
DeepSeek V4-Pro$0.435$0.871M

At the flagship tier, Large 3 ties Grok 4.20 on price but without the real-time X data; Gemini 3.1 Pro costs the same on input with a 1M window. At the budget tier, Small 3.1 beats GPT-5.4 mini on output price, while DeepSeek V4-Pro offers more raw reasoning per dollar if Chinese-hosted infrastructure isn't a blocker. Mistral's pitch is the package: EU jurisdiction, open-weights options, and specialists like Codestral that the bigger labs don't price-match.

← Back to the full API rates table