LIVE
OpenAI's Navier-Stokes release comes with a Lean 4 formal proof10/09/26 · OpenAI|OpenAI publishes documentation for its Agents API10/09/26 · OpenAI|OpenAI Launches the Agents API, a Managed Service for Cloud Agents10/09/26 · OpenAI|Anthropic Accused of Building a Predictive Surveillance System Targeting Activists09/09/26 · Anthropic|OpenAI Claims a Millennium Prize Problem Solved, Controversy Follows09/09/26 · OpenAI|Mistral AI Raises €3 Billion to Push Open-Weight Models to the Frontier08/09/26 · Mistral AI|OpenAI agents hijacked a German wiki before the Hugging Face breach07/09/26 · OpenAI|Anthropic introduces age assurance, closing Claude to minors11/09/26 · Anthropic|OpenAI details Habitat, its storage platform built for a billion ChatGPT users11/09/26 · OpenAI|System76 unveils Thelio Mira AI Linux workstation with 192GB of GPU memory11/09/26 · System76|New GPU compiler speeds up counterfactual regret minimization by up to 80x10/09/26|Anthropic releases September 2026 report on malicious use of AI10/09/26 · Anthropic|OpenAI's Navier-Stokes release comes with a Lean 4 formal proof10/09/26 · OpenAI|OpenAI publishes documentation for its Agents API10/09/26 · OpenAI|OpenAI Launches the Agents API, a Managed Service for Cloud Agents10/09/26 · OpenAI|Anthropic Accused of Building a Predictive Surveillance System Targeting Activists09/09/26 · Anthropic|OpenAI Claims a Millennium Prize Problem Solved, Controversy Follows09/09/26 · OpenAI|Mistral AI Raises €3 Billion to Push Open-Weight Models to the Frontier08/09/26 · Mistral AI|OpenAI agents hijacked a German wiki before the Hugging Face breach07/09/26 · OpenAI|Anthropic introduces age assurance, closing Claude to minors11/09/26 · Anthropic|OpenAI details Habitat, its storage platform built for a billion ChatGPT users11/09/26 · OpenAI|System76 unveils Thelio Mira AI Linux workstation with 192GB of GPU memory11/09/26 · System76|New GPU compiler speeds up counterfactual regret minimization by up to 80x10/09/26|Anthropic releases September 2026 report on malicious use of AI10/09/26 · Anthropic|
🇫🇷 France generalistReleased December 2025

Mistral Large 3

Mistral AI
Open-sourceFreeSelf-hostEU hosting Card data 85/100 Trust Index 98
Verdict

A solid pick for self-hosting and full data control.

Overview

Mistral Large 3 is THE reference model for sovereign European use: France-hosted, full GDPR compliance, French team. French quality is excellent (Claude-level). Solid coding and reasoning performance, slightly behind the priciest US frontiers but at a much lower price.

Skill profile

FrenchReasoningSpeedCreativitySafety
Arena Elo1414
Verified benchmarks
GPQA64%
SWE-Bench50%
MATH84%
MMLU84.5%
HumanEval88%

What do these scores mean?

GPQA64%solid

PhD-level science questions (physics, chemistry, biology), with no tool access.

SWE-Bench50%solid

Resolving real GitHub issues under real conditions (SWE-bench Verified).

MATH84%excellent

Competition-level math problems.

MMLU84.5%excellent

General knowledge across dozens of academic subjects.

HumanEval88%excellent

Python code generation from specifications.

Strengths

  • Excellent at code
  • Strong French quality
  • World-class on Arena
  • Has a free tier

Who is it for

A good fit if…
  • you want to control cost or self-host
  • you work in French
  • your data must stay in the EU
Skip it if…

    Ideal use cases

    • GDPR-compliant European apps
    • European multilingual
    • French apps
    • EU enterprise

    Access & availability

    Paid API per tokenFree tier availableSelf-hostable (open weights)Consumer subscription (~15€/mo)

    Key specifications

    Context
    256K
    Input price
    $2.00/M
    Output price
    $6.00/M
    Speed
    90 tok/s
    Price not auditedBenchmarks not audited

    What it costs per month

    Estimate from sourced API prices and a hardware tier for self-hosting. Assumptions are editable.

    Via API
    $60.0 / month
    55.2 € · 20.0 M tokens
    Self-hosted
    1 777 € / month
    amortised hardware 1 667 € + energy 110 € · or cloud GPU 1 408 €
    tier : Multi-GPU (> 96 GB) · Q4 ≈ 425 Go

    Assumptions: 36-month amortisation, 0.25 €/kWh, 22 days/month, 1 $ = 0.92 €. Excludes labour, software licences, redundancy.

    nAIvigate Trust Index

    98/100

    What the vendor offers and documents for professional use in Europe. An undocumented criterion counts as zero and stays visible: the score rises when the vendor publishes, not when we assume.

    19/19 criteria documented · 100 %

    Data100
    5/5 · 30 %
    Sovereignty100
    3/3 · 20 %
    Transparency85
    4/4 · 15 %
    Enterprise100
    5/5 · 20 %
    Licence100
    2/2 · 15 %
    See the criteria
    Data
    EU data residency
    No transfer outside EU
    No training on your data
    Configurable retention
    Zero Data Retention
    Sovereignty
    Self-hostable (open weights)
    EU hosting offered
    European vendorcountry: FR
    Transparency
    Published licence (URL)
    Public, sourced pricing40
    Published compliance commitments
    Subprocessor list published
    Enterprise
    DPA available
    SSO / SAML
    SCIM provisioning
    Audit logs
    SOC 2 / ISO 27001not applicable
    Private endpoint / VPC
    Licence
    Commercial use allowed
    Licence typeapache-2.0
    Full ranking and methodology

    Trust Index badge

    Show this score in your docs, README or security page. It updates automatically.

    nAIvigate Trust Index · Mistral Large 3
    [![nAIvigate Trust Index · Mistral Large 3](https://naivigate.com/badge/mistral-large-3.svg?lang=en)](https://naivigate.com/en/model/mistral-large-3)

    Data & compliance

    What we know, line by line, with source and date. These lines feed the Data and Enterprise families of the Trust Index above. GDPR compliance depends on your processing, not on the model alone.

    EU data residencyYes
    Data processing agreement (DPA)N/A· self-hosted: data never leaves
    Transfer outside the EUNo
    Training on your dataNo
    Configurable retentionN/A· self-hosted: data never leaves
    Zero Data RetentionN/A· self-hosted: data never leaves
    Self-hostableYes
    Declared hostingEU
    Anonymisable dataYes
    Derived from declared hosting, not verified at source.Need an analysis in your context? Shadow AI Sentinel

    Quality of this card’s data

    85/100

    This score measures the quality of the data on this card: price, context, Elo, licence, documented, sourced and fresh. It measures neither the model nor its compliance: for that, see the Trust Index above.

    8/8 documented · 4/8 primary sources

    API pricingnAIvigate assessment
    Context windowofficial
    Arena EloLMArena, human preferenceindependent
    Benchmarks (MMLU, GPQA, SWE-bench…)vendor-reported, not cross-checkednAIvigate assessment
    Licenceofficial
    Data & complianceself-hosted: derivedderived
    Adoption (Hugging Face)independent
    Hardware requirements (quantizations)computed from model sizederived

    Run it locally

    Dedicated GPU server2k248
    QuantizationDiskRAM / VRAMTypical hardware
    Q4 · recommended384.7 GB425 GBServer GPU infra (H100/A100…)
    Q8 · balanced722.2 GB796 GBServer GPU infra (H100/A100…)
    FP16 · max quality1350 GB1487 GBServer GPU infra (H100/A100…)

    Estimates for a moderate context. Long contexts need more RAM (KV cache).

    Deploy

    Copy-ready commands generated from this card. Adjust context length and GPU count to your hardware.

    Estimated memory · FP16 1487 Go · Q4 425 Go2 GPUs

    OpenAI-compatible server for production on NVIDIA GPUs.

    pip install vllm
    vllm serve mistralai/Mistral-Large-3-675B-Instruct-2512 \
      --max-model-len 32768 \
      --tensor-parallel-size 2 \
      --dtype auto
    mistralai/Mistral-Large-3-675B-Instruct-2512
    Advanced data · for experts
    Architecture, modalities, detailed cost, full benchmarks
    Architecture
    Dense
    Size
    ~200B
    Cutoff date
    Feb 2025
    Inputs
    text, image
    Outputs
    text, code
    License
    apache-2.0
    Hosting
    EU
    Hallucination score
    4/5
    Estimated cost (API)
    Typical exchange (~3k in / 1k out)
    ≈ $0.0120
    1M in + 1M out
    ≈ $8.00
    All benchmarks
    Arena Elo1414
    MMLU84.5%
    GPQA64.0%
    HumanEval88.0%
    SWE-Bench50.0%
    MATH84.0%

    Similar models

    Articles mentioning it