macrostack

AI API price index · cloud platform · read 17 Nov 2026

Azure AI Foundry Models API pricing

69 prices as Azure AI Foundry Models publishes them — 47 text models, 3 embeddings, 2 rerank, 11 images and documents, 4 hardware time — read on its own pricing page on 17 Nov 2026. Its current text models cost $0.30 to $8.00 per million output tokens.

Free tier No free token allowance is printed on the Foundry Models pricing pages. (Azure's general free account credit for new subscriptions is not on these pages.)

What to know before you build on Azure AI Foundry Models

  • Foundry Models is Azure's catalogue of models from makers other than OpenAI, and Microsoft's own; OpenAI's models on Azure have their own page in this index (Azure OpenAI).
  • Prices here are for Global deployments. Data Zone and Regional deployments cost about 10% more: DeepSeek R1 is $1.35 in and $5.40 out Global, $1.485 and $5.94 in a Data Zone.
  • DeepSeek V4.1 Flash is $0.30 in and $1.20 out per million tokens, the same as DeepSeek's own peak price.
  • Phi models run from $0.075 in and $0.30 out (Phi-4-mini) to $0.17 and $0.68 (Phi-3-medium).
  • Grok, Mistral Medium and Cohere's models are resold and ranked only at the maker's own price. The five the index also lists at their maker (Grok 4.3, 4.6 and 4.7, Embed 5 Fast and Embed 5 Pro) cost exactly the same here.
  • Managed compute to host a model yourself is billed per GPU hour: an A100 80GB $3.95, an H100 $7.91, an H200 $9.10.

Against the other 31

Across the index, 390 current text models run from $0.03 to $50.00 per million output tokens — 1667×. Azure AI Foundry Models’s cheapest, Phi-4-mini at $0.30, has 48 cheaper text models elsewhere; its dearest, MAI-Thinking-1, costs $8.00. A cheaper model is a different model, not the same one for less — compare capability before price.

The same model elsewhere

Open-weight models Azure AI Foundry Models sells that other hosts price too — the only like-for-like comparison in the index.

  • Llama 3.3 70B — Novita AI $0.135 / $0.40 · Azure AI Foundry Models $0.71 / $0.71 · Amazon Bedrock $0.72 / $0.72 · Google Vertex AI (Agent Platform) $0.72 / $0.72 · SambaNova Cloud $0.60 / $1.20 — 3.0× apart
  • DeepSeek V4.1 Flash — DeepInfra $0.20 / $0.60 · Azure AI Foundry Models $0.30 / $1.20 · Fireworks AI $0.30 / $1.20 · Google Vertex AI (Agent Platform) $0.30 / $1.20 · Novita AI $0.30 / $1.20 · Together AI $0.30 / $1.20 — 2.0× apart
  • DeepSeek V4 Pro — DeepInfra $1.30 / $2.60 · Novita AI $1.60 / $3.20 · Azure AI Foundry Models $1.74 / $3.48 — 1.3× apart

Input / output per 1M tokens, standard tier.

The same model at its maker

Azure AI Foundry Models resells models other companies make. Here is each one beside its maker's own price, both at the standard global rate and the standard prompt length.

ModelAzure AI Foundry ModelsAt its makerDifference
Grok-3$3.00 / $15.00—Not in the index at its maker
Grok-3 Mini$0.25 / $1.27—Not in the index at its maker
Grok Code Fast 1$0.20 / $1.50—Not in the index at its maker
Grok 4$3.00 / $15.00—Not in the index at its maker
Grok 4 Fast$0.20 / $0.50—Not in the index at its maker
Grok 4.1 Fast$0.20 / $0.50—Not in the index at its maker
Grok 4.2$1.25 / $2.50—Not in the index at its maker
Grok 4.3$1.25 / $2.50$1.25 / $2.50 xAIThe same price
Grok 4.6$2.00 / $6.00$2.00 / $6.00 xAIThe same price
Grok 4.7$2.00 / $6.00$2.00 / $6.00 xAIThe same price
Mistral Medium 3.5$1.50 / $7.50—Not in the index at its maker
Command A$2.50 / $10.00—Not in the index at its maker
Command A Plus$0.80 / $3.20—Not in the index at its maker
Embed v4$0.12—Not in the index at its maker
Embed v5 Fast$0.08$0.08 CohereThe same price
Embed v5 Pro$0.12$0.12 CohereThe same price

Input / output per 1M tokens (embeddings: input only). 5 of 16 cost the same here as at their maker. The index ranks each of these models once, at its maker’s price.

Text models

ModelInputCached inputOutputNote
DeepSeek V3.2$0.58—$1.68Global deployment
DeepSeek V3.2 SP$0.58—$1.68Global deployment
DeepSeek R1$1.35—$5.40Global deployment
DeepSeek V3$1.14—$4.56Global deployment
DeepSeek V3 0324$1.14—$4.56Global deployment
DeepSeek-V3.1$1.23—$4.94Global deployment
DeepSeek V4 Pro$1.74$0.145$3.48Global deployment
DeepSeek V4 Flash$0.19$0.028$0.51Global deployment
DeepSeek V4 Flash 0731$0.44$0.014$1.32Global deployment
DeepSeek V4.1 Flash$0.30$0.006$1.20Global deployment
Grok-3$3.00—$15.00Global deployment · Not ranked: xAI's own model, resold here; the index ranks a maker's model only at its maker's own price.
Grok-3 Mini$0.25—$1.27Global deployment · Not ranked: xAI's own model, resold here; the index ranks a maker's model only at its maker's own price.
Grok Code Fast 1$0.20—$1.50Not ranked: xAI's own model, resold here; the index ranks a maker's model only at its maker's own price.
Grok 4$3.00—$15.00Not ranked: xAI's own model, resold here; the index ranks a maker's model only at its maker's own price.
Grok 4 Fast$0.20—$0.50Not ranked: xAI's own model, resold here; the index ranks a maker's model only at its maker's own price.
Grok 4.1 Fast$0.20—$0.50Global deployment · Not ranked: xAI's own model, resold here; the index ranks a maker's model only at its maker's own price.
Grok 4.2$1.25—$2.50Global deployment · Not ranked: xAI's own model, resold here; the index ranks a maker's model only at its maker's own price.
Grok 4.3$1.25—$2.50Global deployment · Not ranked: the same model xAI sells; ranked once, at xAI's own price.
Grok 4.6$2.00$0.50$6.00Global deployment · Not ranked: the same model xAI sells; ranked once, at xAI's own price.
Grok 4.7$2.00$0.50$6.00Global deployment · Not ranked: the same model xAI sells; ranked once, at xAI's own price.
Grok 4.7$4.00$1.00$12.00Long context: the page's "Grok-4.7 Long Context" price; Global deployment · Not ranked: the same model xAI sells; ranked once, at xAI's own price.
Kimi K2 Thinking$0.60—$2.50Global deployment
Kimi K2.5 Thinking$0.60$0.10$3.00Global deployment
Kimi K2.6 Thinking$0.95$0.16$4.00Global deployment
Kimi K2.7 Code$0.95$0.19$4.00—
Llama 3.3 70B$0.71—$0.71Global deployment
Llama 4 Maverick 17B$0.25—$1.00Global deployment (the page spells this row "Mavrick")
Mistral Medium 3.5$1.50—$7.50Global deployment · Not ranked: Mistral's own model, resold here; the index ranks a maker's model only at its maker's own price.
Command A$2.50—$10.00Global deployment · Not ranked: Cohere's own model, resold here; the index ranks a maker's model only at its maker's own price.
Command A Plus$0.80—$3.20Global deployment · Not ranked: Cohere's own model, resold here; the index ranks a maker's model only at its maker's own price.
MAI-DS-R1$1.35—$5.40Global deployment; Microsoft's post-trained DeepSeek R1
MAI-Thinking-1$2.00$0.20$8.00Global deployment
MAI-Cyber-1 Flash$0.60$0.06$3.50Global deployment
Phi-3-mini-4k-instruct$0.13—$0.524K context
Phi-3-mini-128k-instruct$0.13—$0.52128K context
Phi-3.5-mini-instruct$0.13—$0.52128K context
Phi-3-small-8k-instruct$0.15—$0.608K context
Phi-3-small-128k-instruct$0.15—$0.60128K context
Phi-3-medium-4k-instruct$0.17—$0.684K context
Phi-3-medium-128k-instruct$0.17—$0.68128K context
Phi-3.5-MoE-instruct$0.16—$0.64128K context
Phi-4$0.125—$0.50128K context
Phi-4-mini$0.075—$0.30128K context
Phi-4-multimodal$0.08—$0.32128K context; text and image input
Phi-4-mini-reasoning$0.075—$0.30128K context
Phi-4-reasoning$0.125—$0.5032K context
Phi-4-reasoning-plus$0.125—$0.5032K context

Embeddings

ModelInputCached inputNote
Embed v4$0.12—text input; image input $0.47 per 1M tokens; Global deployment · Not ranked: Cohere's own model, resold here; the index ranks a maker's model only at its maker's own price.
Embed v5 Fast$0.08—text input; image input $0.4 per 1M tokens; Global deployment · Not ranked: the same model Cohere sells; ranked once, at Cohere's own price.
Embed v5 Pro$0.12—text input; image input $0.4 per 1M tokens; Global deployment · Not ranked: the same model Cohere sells; ranked once, at Cohere's own price.

Rerank

ModelPriceNote
Rerank v4 Fast$2.00 per 1K searchesGlobal deployment
Rerank v4 Pro$2.50 per 1K searchesGlobal deployment

Images and documents

ModelPriceNote
mistral-document-ai-2505$3.00 per 1K pagesdocument OCR
Mistral OCR$1.00 per 1K pagesdocument OCR
Mistral-document-ai-2512$3.00 per 1K pagesdocument OCR
Mistral OCR 2512$2.00 per 1K pagesdocument OCR
Mistral OCR 4 Annotated$5.00 per 1K pagesdocument OCR
Mistral OCR 4$4.00 per 1K pagesdocument OCR
MAI-Image-2$33.00 per 1M tokensimage output tokens; input $5.00 per 1M tokens
MAI-Image-2-Efficient$19.50 per 1M tokensimage output tokens; input $5.00 per 1M tokens
MAI-Image-2.5$47.00 per 1M tokensimage output tokens; input text $5.00, image $8.00 per 1M tokens
MAI-Image-2.5 Flash$19.50 per 1M tokensimage output tokens; input text $1.75, image $1.75 per 1M tokens
MAI-Image-2.5 Pro$106.00 per 1M tokensimage output tokens; input text $5.00, image $8.00 per 1M tokens

Hardware time

ModelPriceNote
A100 80GB (managed compute)$3.95 per hourper GPU compute hour, to host a model yourself
H100 (managed compute)$7.91 per hourper GPU compute hour, to host a model yourself
H200 (managed compute)$9.10 per hourper GPU compute hour, to host a model yourself
MI300 (managed compute)$7.91 per hourper GPU compute hour, to host a model yourself

Other prices

ModelPriceNote
Microsoft-Decision-1$0.042 per 1M tokensinput tokens; no output price printed
Phi-4-multimodal (audio input)$4.00 per 1M tokensaudio input tokens; text output $0.32 per 1M tokens

33 of these 69 prices are ranked in the index; the rest say why in their note. Text and embeddings per 1M tokens unless the row says otherwise.

Discounts and surcharges, as published

  • Global deployments are the cheapest. Data Zone and Regional deployments cost about 10% more (DeepSeek R1: $1.35 in and $5.40 out Global, $1.485 and $5.94 Data Zone or Regional).
  • Provisioned throughput: DeepSeek R1, V3, V3 0324 and MAI-DS-R1 at $1.00 per PTU-hour or $260 per PTU-month, 100 PTU minimum.
  • Phi models can be fine-tuned at $3.00 per 1M training tokens plus $0.80 an hour to host, with usage at the same token price.

Where these numbers come from

Read on azure.microsoft.com/en-us/pricing/details/microsoft-foundry on 17 Nov 2026, checked against 7 more of Azure AI Foundry Models’s own pages. Macrostack measures nothing and is paid by no provider on this index. Prices change; the date is the claim.

The other providers in the index

Questions

How much does the Azure AI Foundry Models API cost?
On 17 Nov 2026, Azure AI Foundry Models's current text models cost $0.30 to $8.00 per million output tokens at the standard tier (Phi-4-mini to MAI-Thinking-1). The full list, with input and cached-input prices, is on this page.
Is Azure AI Foundry Models cheaper than buying from the model maker directly?
For 5 of 16 models the price is the same: on 17 Nov 2026 Grok 4.3 cost $1.25 in and $2.50 out per million tokens on Azure AI Foundry Models, against $1.25 and $2.50 at xAI; Grok 4.6 cost $2.00 in and $6.00 out per million tokens on Azure AI Foundry Models, against $2.00 and $6.00 at xAI; Grok 4.7 cost $2.00 in and $6.00 out per million tokens on Azure AI Foundry Models, against $2.00 and $6.00 at xAI. The reasons to choose one over the other are billing, data residency, compliance and the rest of the cloud you already use, not the token price.
Does Azure AI Foundry Models have a free tier?
What Azure AI Foundry Models's own pages say: No free token allowance is printed on the Foundry Models pricing pages. (Azure's general free account credit for new subscriptions is not on these pages.)
Does Azure AI Foundry Models offer discounts?
Global deployments are the cheapest. Data Zone and Regional deployments cost about 10% more (DeepSeek R1: $1.35 in and $5.40 out Global, $1.485 and $5.94 Data Zone or Regional). Provisioned throughput: DeepSeek R1, V3, V3 0324 and MAI-DS-R1 at $1.00 per PTU-hour or $260 per PTU-month, 100 PTU minimum. Phi models can be fine-tuned at $3.00 per 1M training tokens plus $0.80 an hour to host, with usage at the same token price.
The Macrostack brief

New swaps, worth your inbox.

A short, occasional email when we add a high-intent alternative or ship a new head-to-head. No spam, no selling your address — unsubscribe in one click.