Skip to main content
Watch: I Built an AI App Without Leaving Databricks

Fine-tuning - verified

Microsoft Foundry fine-tuning: status, methods and prices

Offered. 3 offerings recorded: SFT, DPO and RFT. Training is priced per 1M training tokens and per training hour, in Microsoft Foundry's own units. Weight export: Not stated.

Recorded as: Microsoft Foundry (Azure AI Foundry / Azure OpenAI).

Read from learn.microsoft.com, azure.microsoft.com. Compare every fine-tuning API

Status
Offered
Offerings
3
Methods
SFT, DPO, RFT
Export weights
Not stated

Status

Is Microsoft Foundry fine-tuning still available?

Offered

Offered

All 3 offerings

Prices are in the provider's own unit, as written. Not stated means the page we checked does not say.
SFTGA, some models preview

Managed supervised fine-tuning

Supervised fine-tuning

InferredThe pricing page renders prices for a default region and does not state which one. The Regional, Global and Developer prices shown are for that unstated region.

Status

GA for OpenAI GPT-4o/4.1 family; several open models marked preview

“gpt-oss-120b, (preview) · Managed: [yes] (SFT, RFT), Interactive: [yes]”
learn.microsoft.com
Tunable base models
  • gpt-4o-mini (2024-07-18)
  • gpt-4o (2024-08-06)
  • gpt-4.1 (2025-04-14)
  • gpt-4.1-mini (2025-04-14)
  • gpt-4.1-nano (2025-04-14)
  • gpt-oss-20b
  • gpt-oss-120b (preview)
  • Llama-3.3-70B-Instruct
  • Qwen3-32B
  • Qwen3.6-35B-A3B (preview)
  • Qwen3.8-27B (preview)
  • Muse-Glimmer-30B (preview)
  • Ministral-3B (2411)
  • MedImageInsight-Premium (preview)
  • CXRReportgen-Premium (preview)

Full list from the supported-models table; four representative rows quoted.

“gpt-4.1, (2025-04-14) · Managed: [yes] (SFT, DPO), Interactive: [no]”
learn.microsoft.com
“gpt-oss-20b · Managed: [yes] (SFT), Interactive: [yes]”
learn.microsoft.com
“Llama-3.3-70B-Instruct · Managed: [yes] (SFT), Interactive: [no]”
learn.microsoft.com
“Qwen3-32B · Managed: [yes] (SFT), Interactive: [no]”
learn.microsoft.com
Training price
Managed supervised fine-tuning training prices in Microsoft Foundry (Azure AI Foundry / Azure OpenAI)'s own units
ModelPriceUnitQuote
GPT-4.1 (Regional)$27.5per 1M training tokens“Training: $27.5/1M tokens, Hosting: $1.7/hour”azure.microsoft.com
GPT-4.1 (Global)$25per 1M training tokens“Global · Input: $2/1M tokens, Cached Input: $0.50/1M tokens, Output: $8/1M tokens, Training: $25/1M tokens, Hosting: $1.7/hour”azure.microsoft.com
GPT-4.1 (Developer)$12.50per 1M training tokens“Developer · Input: $2/1M tokens, Cached Input: $0.50/1M tokens, Output: $8/1M tokens, Training: $12.50/1M tokens”azure.microsoft.com
GPT-4.1-mini (Global)$5per 1M training tokens“Global · Input: $0.40/1M tokens, Cached Input: $0.10/1M tokens, Output: $1.60/1M tokens, Training: $5/1M tokens, Hosting: $1.7/hour”azure.microsoft.com
GPT-4.1-nano (Global)$1.5per 1M training tokens“Global · Input: $0.10/1M tokens, Cached Input: $0.025/1M tokens, Output: $0.40/1M tokens, Training: $1.5/1M tokens, Hosting: $1.7/hour”azure.microsoft.com
GPT OSS 20B (Regional)$3.60per 1M training tokens“GPT OSS 20B · Regional · Input: $0.07/1M tokens, Output: $0.3/1M tokens, Training: $3.60/1M tokens, Hosting: $0.3/hour”azure.microsoft.com
Tuned-model inference

  • GPT-4.1 (Global): $2 input / $8 output per 1M tokens + $1.7/hour hosting (per 1M tokens + per hour hosting)

    “Global · Input: $2/1M tokens, Cached Input: $0.50/1M tokens, Output: $8/1M tokens, Training: $25/1M tokens, Hosting: $1.7/hour”
    azure.microsoft.com

  • GPT OSS 20B (Regional): $0.07 input / $0.3 output per 1M tokens + $0.3/hour hosting (per 1M tokens + per hour hosting)

    “GPT OSS 20B · Regional · Input: $0.07/1M tokens, Output: $0.3/1M tokens, Training: $3.60/1M tokens, Hosting: $0.3/hour”
    azure.microsoft.com
Hosting requirement

Hourly hosting fee for Standard/Regional/Global deployments (e.g. $1.7/hour for GPT-4.1 family); Developer deployments have no hourly fee

“Runs inference globally with per-token billing and no hourly hosting fee or data-residency guarantee. For evaluation only; lasts 24 hours and has no availability SLA.”
learn.microsoft.com
Export weights

Not stated

Data limits
Not stated
Your training data

Global training copies data and weights to the training region

“Data and weights are copied to the region where training occurs.”
learn.microsoft.com
Notes

GPT-4o and GPT-4o-mini fine-tuning rows show N/A in the rendered pricing region; they remain in the supported-models table.

DPONot stated

Managed preference fine-tuning (DPO)

Direct preference optimization (preference tuning)

Status
Not stated
Tunable base models
  • gpt-4o (2024-08-06)
  • gpt-4.1 (2025-04-14)
  • gpt-4.1-mini (2025-04-14)
  • gpt-4.1-nano (2025-04-14)
“gpt-4o, (2024-08-06) · Managed: [yes] (SFT, DPO), Interactive: [no]”
learn.microsoft.com
“gpt-4.1, (2025-04-14) · Managed: [yes] (SFT, DPO), Interactive: [no]”
learn.microsoft.com
“gpt-4.1-mini, (2025-04-14) · Managed: [yes] (SFT, DPO), Interactive: [no]”
learn.microsoft.com
“gpt-4.1-nano, (2025-04-14) · Managed: [yes] (SFT, DPO), Interactive: [no]”
learn.microsoft.com
Training price

Not stated

Pricing page has no DPO-specific training row.

Tuned-model inference
Not stated
Hosting requirement

Hourly hosting fee for Standard/Regional/Global deployments (e.g. $1.7/hour for GPT-4.1 family); Developer deployments have no hourly fee

“Runs inference globally with per-token billing and no hourly hosting fee or data-residency guarantee. For evaluation only; lasts 24 hours and has no availability SLA.”
learn.microsoft.com
Export weights

Not stated

Data limits
Not stated
RFTGA, some models preview

Managed reinforcement fine-tuning

Reinforcement fine-tuning

InferredThe pricing page renders prices for a default region and does not state which one. The Regional, Global and Developer prices shown are for that unstated region.

Status

GA for o4-mini; GPT-5 RFT invitation-only; open-model RFT preview

“GPT-5 reinforcement fine-tuning requires an invitation. Contact your Microsoft account team for enrollment.”
learn.microsoft.com
Tunable base models
  • o4-mini (2025-04-16)
  • gpt-5 (2025-08-07, invitation only)
  • MAI-Code-1.1-Flash (preview)
  • Muse-Glimmer-30B (preview)
  • Qwen3.6-35B-A3B (preview)
  • Qwen3.8-27B (preview)
  • gpt-oss-120b (preview)
“o4-mini, (2025-04-16) · Managed: [yes] (RFT), Interactive: [no]”
learn.microsoft.com
“GPT-5 reinforcement fine-tuning requires an invitation. Contact your Microsoft account team for enrollment.”
learn.microsoft.com
“MAI-Code-1.1-Flash, (preview) · Managed: [yes] (RFT), Interactive: [no]”
learn.microsoft.com
Training price
Managed reinforcement fine-tuning training prices in Microsoft Foundry (Azure AI Foundry / Azure OpenAI)'s own units
ModelPriceUnitQuote
o4-mini (Global)$100per hour of training“Global · Input: $1.10/1M tokens, Output: $4.40/1M tokens, Training: $100/hour, Hosting: $1.7/hour,”azure.microsoft.com
o4-mini (Regional)$110per hour of training“Training: $110/hour, Hosting: $1.7/hour”azure.microsoft.com
o4-mini (Developer)$50per hour of training“Developer · Input: $1.10/1M tokens, Cached Input: $0.275/1M tokens, Output: $4.40/1M tokens, Training: $50/hour”azure.microsoft.com
Tuned-model inference

  • o4-mini RFT (Global): $1.10 input / $4.40 output per 1M tokens + $1.7/hour hosting (per 1M tokens + per hour hosting)

    “Global · Input: $1.10/1M tokens, Output: $4.40/1M tokens, Training: $100/hour, Hosting: $1.7/hour,”
    azure.microsoft.com
Hosting requirement

Hourly hosting fee for Standard/Regional/Global deployments (e.g. $1.7/hour for GPT-4.1 family); Developer deployments have no hourly fee

“Runs inference globally with per-token billing and no hourly hosting fee or data-residency guarantee. For evaluation only; lasts 24 hours and has no availability SLA.”
learn.microsoft.com
Export weights

Not stated

Data limits
Not stated
Notes

Grader tokens billed separately (GPT-4o and o3-mini grader rates listed on the pricing page).

Research notes

Managed fine-tuning supports SFT, DPO and RFT (model-specific); distillation is done via teacher-generated SFT data. Training types: Standard (regional), Data zone (US only), Global (cheaper, copies data and weights), Developer (preemptible, cheapest). Pricing page values below are from whatever region the page rendered by default; the page notes 'Pricing is not available in the selected region' for some rows and GPT-4o / GPT-4o-mini rows show N/A. An interactive training (preview) approach also exists.

Plain language

The methods, explained

SFTSupervised fine-tuning
You show the model example prompts with the answers you want, and it learns to answer like that. The default method almost everywhere.
DPODirect preference optimization (preference tuning)
You give pairs of answers to the same prompt, one preferred and one not, and the model learns to lean toward the preferred kind. Good for tone and style.
RFTReinforcement fine-tuning
The model attempts tasks and a grader (your code, or another model) scores each attempt; training pushes it toward higher-scoring answers. Often billed by training hour, plus the grader's tokens.
Vision SFTVision fine-tuning
Supervised fine-tuning where the examples include images, so the model gets better at a visual task.
DistillationDistillation
A large teacher model writes answers that are used to fine-tune a smaller, cheaper student model. You pay for the teacher's calls as well as the training.
LoRALoRA (adapter) fine-tuning
Trains a small set of extra weights, an adapter, on top of the frozen base model. Cheaper and faster than full fine-tuning; the adapter is served alongside the base model.
FullFull-parameter fine-tuning
Updates every weight in the model. Costs more than LoRA and produces a complete copy of the model.
CPTContinued pre-training
Keeps training the model on a large body of raw, unlabeled text from your domain, so it absorbs vocabulary and facts before any instruction tuning.

FAQ

Does Microsoft Foundry offer fine-tuning?

Offered. 3 offerings: SFT, DPO and RFT.

How much does Microsoft Foundry fine-tuning cost?

Per 1M training tokens: GPT-4.1 (Regional) $27.5, GPT-4.1 (Global) $25, GPT-4.1 (Developer) $12.50, GPT-4.1-mini (Global) $5 and 2 more. Per training hour: o4-mini (Global) $100, o4-mini (Regional) $110, o4-mini (Developer) $50. Not stated for Managed preference fine-tuning (DPO). Prices are in Microsoft Foundry's own units and are not converted.

Can you download a model you fine-tune on Microsoft Foundry?

Not stated. The pages we checked do not say whether tuned weights can be exported.

Sources

Last verified . This is a directory of what each provider's own pages say, not a quality ranking. Prices are copied in the provider's own unit and never converted, so rows in different units are not comparable.

Newsletter

Know when a provider changes its fine-tuning terms

One email per week. New model releases, deprecations and pricing changes, and what they mean for what you build.