Best Fine Tuning Companies

Vendor profileChecked Nov 8, 2026

Microsoft Foundry fine-tuning

Microsoft Corporation (formerly Azure AI Foundry) · ai.azure.com · Last checked:

  • Managed fine-tuning API
  • Cloud platform fine-tuning
  • Reinforcement fine-tuning

Company status

Active. Business line of Microsoft Azure; parent Microsoft Corporation is publicly listed.

Sources 1, 2, 3, 4 · checked 2026-11-08

Capabilities

  • Managed Training: SFT, DPO, and RFT depending on model; distillation via teacher-generated SFT data
  • Programmed Training (preview, approval required): Python training loops with custom losses, rewards, and rollouts
  • Fine-tunable models include gpt-4.1 family, gpt-4o, o4-mini (RFT), gpt-oss-20b/120b, Llama-3.3-70B-Instruct, Qwen3-32B, Ministral-3B
  • Training types: Standard (regional residency), Data zone (US), Global, Developer (preemptible)
  • Deployment: serverless, managed compute (preview), or Fireworks on Foundry (preview)

Sources 1, 2, 3, 4 · checked 2026-11-08

Who it tends to fit

  • Teams that need to keep fine-tuning OpenAI models (gpt-4.1, o4-mini) inside Azure
  • Buyers needing regional data residency for training

Sources 1, 2, 3, 4 · checked 2026-11-08

Pricing

Per Azure retail prices (East US 2): gpt-4.1 training $0.0275 per 1K tokens regional, $0.025 global, $0.0125 developer; gpt-4.1-mini training $0.0055 per 1K tokens regional, $0.005 global; fine-tuned model hosting $1.70 per hour. Rates vary by model, region, and training type.

Sources 1, 2, 3, 4 · checked 2026-11-08

Published certifications

  • FedRAMP High (Azure OpenAI in Azure FedRAMP audit scope)

Microsoft publishes compliance scope per Azure service in the Service Trust Portal; confirm SOC 2, ISO 27001, and HIPAA scope for Foundry fine-tuning there.

Listed as stated by the vendor. Scope can differ by product or subsidiary; request current reports.

Sources 1, 2, 3, 4 · checked 2026-11-08

Limitations to check

  • Method, training type, and deployment option support differ per model and region
  • GPT-5 RFT requires an invitation; Programmed Training requires preview approval
  • Global and Developer training do not guarantee data residency
  • Pricing page loads rates dynamically; confirm your region in the Azure pricing calculator

Sources 1, 2, 3, 4 · checked 2026-11-08

Sources

  1. Managed vs Programmed Training, SFT/DPO/RFT, training types, deployment options, supported models tablelearn.microsoft.com/en-us/azure/foundry/fine-tuning/overviewChecked 2026-11-08
  2. SFT job workflow and deploying fine-tuned models in the Foundry portallearn.microsoft.com/en-us/azure/foundry/openai/how-to/fine-tuningChecked 2026-11-08
  3. Azure Retail Prices API rows for gpt-4.1 / gpt-4.1-mini FT training and hosting (East US 2)prices.azure.com/api/retail/prices?%24filter=serviceName%20eq%20%27Foundry%20Models%27%20and%20armRegionName%20eq%20%27eastus2%27%20and%20contains(meterName,%274.1%27)Checked 2026-11-08
  4. Azure OpenAI listed with FedRAMP High in Azure audit scopelearn.microsoft.com/en-us/azure/azure-government/compliance/azure-services-in-fedramp-auditscopeChecked 2026-11-08

Report a correction: [email protected]. Facts are re-checked at least every 90 days.