Skip to content
NEWSee what changed →
Real-world adoption

Open models in production

Publicly reported deployments of open-weight language models, grouped by sector. Every story links to its source; new ones are picked up from the news and reviewed before they appear here.

Documented case studies

  • Confidential-computing LLM service for regulated customers

    Edgeless Systems · Privatemode AI

    Built an end-to-end encrypted LLM service for healthcare and banking clients whose data rules ruled out ordinary cloud AI. Chose Llama because it was open, multilingual and self-hostable, and ran a quantised INT4 build to balance quality against compute.

    Models
    Llama 3.1 70B → Llama 3.3 70B (AWQ INT4)
    Deployment
    Self-hosted in confidential-computing enclaves; INT4 quantised weights
    Source: Meta — Llama case study ↗
  • Japan / globalFinancial services

    Firm-wide generative AI on open models

    Nomura

    The investment bank set out to make generative AI available across the firm and standardised on Llama models, citing faster innovation, transparency, bias guardrails and strong results on summarisation and code generation.

    Models
    Llama (Meta)
    Deployment
    Managed hosting on Amazon Bedrock
    Source: AWS customer case study ↗
  • United StatesFintech & payments

    Customer support systems built on an open model

    Block · Cash App

    Block integrated Llama into the support systems behind Cash App. Because the weights are open, the team can experiment and customise per use case while keeping customer data private.

    Models
    Llama (Meta)
    Deployment
    Customised open model inside Block's own systems
    Source: Meta AI blog ↗
  • Automated hospital discharge summaries

    Instituto de Inteligência Artificial na Saúde · NoHarm

    The NoHarm Summary Discharge project drafts and improves patient discharge summaries to cut administrative load, reduce errors and give patients complete post-hospital instructions — one of the highest-value, highest-risk hospital use cases in the playbook.

    Models
    Llama (Meta)
    Deployment
    Built with Llama for clinical documentation
    Source: Meta — Llama AI innovation ↗
  • United Arab EmiratesHealthcare & research

    Bilingual Arabic–English biomedical assistant

    MBZUAI · BiMediX2

    A bilingual expert system providing virtual medical advice and diagnostic support in English and Arabic, including medical-image queries and spoken interaction — an example of open models serving languages that closed APIs cover poorly.

    Models
    Llama (Meta), Seamless-M4T
    Deployment
    Research deployment on open weights with speech models
    Source: Meta — Llama AI innovation ↗
  • US · EU · Japan · Korea · NATO-associatedGovernment & public sector

    Open models for sovereign and national-security use

    US, EU and allied governments

    Meta extended Llama access to US agencies and contractors and later to governments including France, Germany, Italy, Japan and South Korea, noting that open weights can be downloaded and deployed in secure environments at various classification levels without sending data to third-party providers.

    Models
    Llama (Meta)
    Deployment
    Self-hosted in classified / sovereign environments
    Source: Engadget (via Yahoo Tech) ↗
  • France / 160 countriesLogistics & shipping

    Internal assistant for 155,000+ employees

    CMA CGM · MAIA

    The shipping and logistics group deployed MAIA, an internal personal assistant powered by Mistral, to raise workforce productivity across operations in 160 countries.

    Models
    Mistral models
    Deployment
    Enterprise-wide assistant on Mistral models
    Source: Mistral AI — Solutions ↗
  • France / globalInsurance & banking

    Secure AI for 140,000+ insurance staff and a global bank

    AXA and BNP Paribas

    AXA gives 140,000+ employees secure text-generation and analysis capabilities on Mistral models, while BNP Paribas uses them across global markets, sales and customer support under strict compliance and security standards.

    Models
    Mistral models
    Deployment
    Enterprise deployments with compliance controls
    Source: Mistral AI — Solutions ↗
  • Evidence-based medication recommendations for 300+ hospitals

    Synapse Medicine

    Synapse Medicine's clinical-decision platform, serving more than 300 hospitals, runs evidence-based recommendations on Mistral models under HIPAA-aligned controls.

    Models
    Mistral models
    Deployment
    Clinical decision support platform
    Source: Mistral AI — Solutions ↗
  • Learning outcomes and student support at national scale

    India — Ministry of Skill Development and Entrepreneurship

    The ministry is building on Llama to improve learning outcomes and student support, one of several public-sector programmes that chose open weights for cost, customisation and data control.

    Models
    Llama (Meta)
    Deployment
    Government programme on open weights
    Source: Meta AI blog ↗