For Developers/Models/Mistral Large 3
Open SourceDeprecatedMistral AIReleased 2025-12

Mistral Large 3

Mistral Large 3, the previous open-weight Mistral flagship, now deprecated in favor of Mistral Large 4

Mistral AI has replaced this model with Mistral Large 4.

Context window

256K

Input / 1M tokens

$0.50

Output / 1M tokens

$1.50

Provider

Mistral AI

Mistral's docs list Mistral Large 3 at $0.50 per 1M input tokens and $1.50 per 1M output tokens; the Batch API halves that. Mistral deprecated the model on October 31, 2026 and will retire it on April 30, 2027, with Mistral Large 4 ($0.70/$2.10) as the replacement. · Data verified 2026-11-08

This page covers Mistral Large 3 (API names mistral-large-2512 and mistral-large-latest, December 2025), Mistral AI's previous open-weight flagship: a multimodal Mixture-of-Experts model with 675B total and 41B active parameters, a 256K-token context window and an Apache 2.0 license. Mistral Large 4 replaced it: Mistral deprecated Large 3 on October 31, 2026 and will retire it from the API on April 30, 2027. Until then it stays in the API at $0.50/$1.50 per million tokens, a little cheaper than Large 4, and its weights remain on Hugging Face for self-hosting.

Capability index

Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.

Coding
82
Reasoning
85
Math
80
Multimodal
65
Long context
70
Speed
70
Cost efficiency
66

How to access it

Call the Mistral API with model id 'mistral-large-2512' (Mistral's docs still list 'mistral-large-latest' as an alias for it, but the model is deprecated and retires on April 30, 2027), or use it through OpenRouter or Vercel AI Gateway. Open weights are on Hugging Face (mistralai/Mistral-Large-3-675B-Instruct-2512, FP8, with BF16 and NVFP4 variants) for self-hosting.

Strengths

  • ✓General-purpose reasoning and multilingual performance at a lower API price than Mistral Large 4
  • ✓Apache 2.0 open weights for self-hosting and customization
  • ✓Multimodal (handles text and images)
  • ✓Strong at code, JSON, and structured output across 80+ programming languages
  • ✓Broad multilingual coverage (French, German, Spanish, Italian, and many more)

Best for developers who...

Existing Mistral Large 3 integrations until the April 30, 2027 retirementSelf-hosted open-weight deployments that do not need a 1T-parameter modelMultilingual reasoning and generationStructured/JSON output and coding

When to choose it (and when not to)

Reach for Mistral Large 3 when...

  • →You already run Mistral Large 3 and need time to migrate before April 30, 2027
  • →You want to self-host an Apache 2.0 model that is smaller than Mistral Large 4
  • →You want the slightly lower API price ($0.50/$1.50 vs $0.70/$2.10 per 1M tokens)
  • →You need the option to self-host for data sovereignty

Look elsewhere if...

  • ✕You are starting a new API integration (Mistral recommends Mistral Large 4)
  • ✕You need more than 256K tokens of context (Mistral Large 4 offers 1M)
  • ✕You need audio or video input
  • ✕You need the strongest Mistral model available

How to use it

  • ›Use the system message to fix role, language, and output schema
  • ›Request explicit JSON and provide a schema for reliable structured output
  • ›For multilingual tasks, state the target language clearly
  • ›Break complex reasoning into steps or ask for a plan before the answer

Quickstart

Python
from mistralai import Mistral

client = Mistral(api_key="YOUR_MISTRAL_API_KEY")

resp = client.chat.complete(
    model="mistral-large-2512",
    messages=[{"role": "user", "content": "Explain the CAP theorem in three sentences."}],
)
print(resp.choices[0].message.content)

Requires the official 'mistralai' Python SDK (pip install mistralai). Uses the pinned Mistral Large 3 id; use 'mistral-large-4' for new work, since Large 3 retires on April 30, 2027.

API model id: mistral-large-2512

Compare Mistral Large 3

Compare Mistral Large 3 with any other model

Build a comparison →
All model comparisons →

Learn the concepts