Agenticum · Thoughts · 7 min read

Azure OpenAI pricing in the UAE and Saudi Arabia: per model in USD, AED and SAR

Azure OpenAI prices are listed in US dollars, with token-based charges or hourly charges per deployed provisioned throughput unit (PTU). These are the rates Microsoft's Retail Prices API lists for UAE North, converted to dirhams and riyals at the fixed pegs, with notes on what pushes a Gulf company's bill up or down.

Nikolay Zuykov · Agenticum · 3 November 2026

Azure OpenAI pricing in one paragraph

Azure OpenAI offers several billing options. Standard deployments charge per token. Provisioned throughput charges per PTU per hour for the capacity you deploy, even when that capacity handles no requests. Batch and Flex charge per token at half the Standard rate, and in return you accept slower turnaround. Prices are set in US dollars. Every price in this article comes from Microsoft's Retail Prices API (prices.azure.com) for the UAE North region, unless a US East 2 listing is named. We convert at two fixed pegs: 3.6725 dirhams per dollar, as listed by the Central Bank of the UAE, and 3.75 riyals per dollar, the official rate the Saudi Central Bank maintains. Microsoft says its page prices are estimates and not price quotes.

Price per 1M tokens for current models (GPT-6.1, GPT-6, GPT-5.5, GPT-5.4)

All rows are Global Standard as listed for UAE North. For GPT-6.x and GPT-5.5 we show the short-context tier. AED and SAR are rounded to two decimals.

Model	Input USD	Input AED	Input SAR	Output USD	Output AED	Output SAR
GPT-6.1 Sol	2.00	7.35	7.50	10.00	36.73	37.50
GPT-6.1 Terra	1.00	3.67	3.75	5.00	18.36	18.75
GPT-6 Astra	10.00	36.73	37.50	50.00	183.63	187.50
GPT-6 Luna	0.10	0.37	0.38	0.50	1.84	1.88
GPT-5.5	5.00	18.36	18.75	30.00	110.18	112.50
GPT-5.4	2.50	9.18	9.38	15.00	55.09	56.25
GPT-5.4 mini	0.75	2.75	2.81	4.50	16.53	16.88
GPT-5.4 nano	0.20	0.73	0.75	1.25	4.59	4.69

Cached input for GPT-6.1 Sol is $0.10 per 1M tokens (AED 0.37, SAR 0.38), which is one twentieth (5%) of its normal input rate of $2.00. This article does not cover cached rates for the other models, so look up the cached meter for your model in the API.

A price listing for UAE North does not prove that you can deploy a model there. Microsoft's availability table lists GPT-6.1 Sol (version 2026-09-29) on Global Standard and Global Provisioned in UAE North, and GPT-6.1 Terra (version 2026-10-26) on Global Standard. For any other model, check that table before you budget for it.

Global vs Data Zone vs Regional: which one a UAE or KSA company can actually use

For most workloads Microsoft recommends starting with Global Standard, because new models launch there first, it has the lowest price and it covers the most regions. Global deployment also affects where inference data can be processed. On Global types, inference data may be processed in any Azure region. Data stored at rest stays in the designated Azure geography. If your compliance team needs inference to stay inside a specific geography, Global does not promise that, and the Data Zone and Regional rates below cost more than Global Standard.

Deployment type	Where inference is processed	Price compared with Global	Listing used
Global Standard	May be any Azure region; data at rest stays in your geography	Base price	UAE North
Data Zone Standard	Check Microsoft's deployment types page	About 10% higher: GPT-5.4 input is $2.75 against $2.50	US East 2
Regional Standard	Check Microsoft's deployment types page	About 21% higher: GPT-4o (2024-11-20) input is $0.003025 per 1K tokens against $0.0025 on Global	UAE North

Regional pricing in UAE North is also higher than regional pricing in the US. The same GPT-4o regional input meter costs $0.00275 per 1K tokens in US East 2, so the UAE North rate is 10% higher. Our sources do not confirm whether any Data Zone covers the Middle East, or whether Azure OpenAI is offered in a region in Saudi Arabia. Before a company in the UAE or KSA plans around either one, it should ask Microsoft or its Azure partner which region and deployment type its contract will actually use.

How to pay less: Batch, Flex, cached input

Batch. Global Batch runs groups of requests asynchronously on a separate quota, with a 24-hour target turnaround, for 50% less than Global Standard. For GPT-5.4 the Batch rates are $1.25 input and $7.50 output per 1M tokens (AED 4.59 and 27.54, SAR 4.69 and 28.13).

Flex. Flex processing is in preview. It costs 50% less than Standard and is meant for workloads that can tolerate slower responses and occasional resource unavailability. GPT-6.1 Sol input on Flex is listed at $1.00 per 1M tokens (AED 3.67, SAR 3.75). Do not route user-facing traffic through Flex if users cannot wait for an answer.

Cached input. On GPT-6.1 Sol, cached input tokens cost one twentieth of the normal input rate: $0.10 against $2.00 per 1M tokens. Example: in a month with 50M input tokens and 10M output tokens, suppose 40M of the input tokens are a repeated prompt prefix billed at the cached rate. The bill is 10M × $2.00 = $20, plus 40M × $0.10 = $4, plus 10M × $10.00 = $100. That totals $124 (AED 455.39, SAR 465.00), compared with $200 without caching. Ask Microsoft how your prompts qualify for the cached rate before you count on it.

Provisioned throughput (PTU): hourly rates and reservations

PTU type	USD per PTU per hour	AED	SAR	Listing
Global	1.00	3.67	3.75	UAE North
Data Zone	1.10	4.04	4.13	US East 2
Regional	2.36	8.67	8.85	UAE North, effective September 2026
Regional	2.00	7.35	7.50	US East 2

A regional PTU in UAE North costs 18% more than one in US East 2, and 2.36 times as much as a Global PTU. You are billed every hour for the capacity you deploy, whether or not it serves requests. A provisioned deployment cannot be paused: billing stops only when you delete it. Example arithmetic for one PTU over a 30-day month (720 hours): Global comes to $720 (AED 2,644.20, SAR 2,700.00) and Regional in UAE North to $1,699.20 (AED 6,240.31, SAR 6,372.00). Multiply by the number of PTUs you deploy.

Reservations commit you to a fixed number of PTUs for one month or one year at a discounted effective hourly rate. The size of the discount depends on model family and term, so confirm the figure for your model and term with your account team. A reservation does not reserve capacity. Microsoft's advice is to create the deployment first, confirm the capacity is there, and only then buy the reservation.

Worked monthly cost example in AED and SAR

Example workload: 50M input tokens and 10M output tokens a month, with no caching. Steps for GPT-6.1 Sol on Global Standard:

Input: 50 × $2.00 = $100.
Output: 10 × $10.00 = $100.
Total: $200.
Dirhams: $200 × 3.6725 = AED 734.50. Riyals: $200 × 3.75 = SAR 750.00.

The same method gives these totals:

Model and type	USD	AED	SAR
GPT-6.1 Sol, Global Standard	200.00	734.50	750.00
GPT-6.1 Sol, Flex (output computed at the 50% discount)	100.00	367.25	375.00
GPT-6.1 Terra, Global Standard	100.00	367.25	375.00
GPT-5.4 mini, Global Standard	82.50	302.98	309.38
GPT-5.4, Global Standard	275.00	1,009.94	1,031.25
GPT-5.4, Global Batch	137.50	504.97	515.63

We show Flex for GPT-6.1 Sol because its Flex input price is listed, and Batch for GPT-5.4 because both of its Batch meters are listed. Before you plan on Batch or Flex for any other model, check that the model supports them. To rerun the table with your own volumes, replace 50 and 10 with your monthly input and output in millions of tokens.

Checklist before you sign: five questions to ask

Must inference stay in a specific geography? Global types may process inference in any Azure region. If that is not acceptable, ask which Data Zone or regional options your region has and what they cost. Data Zone was about 10% above Global for GPT-5.4 input in the US East 2 listing, and regional in UAE North was about 21% above Global for GPT-4o input.
Which context tier will your prompts be billed on? The GPT-6.x and GPT-5.5 prices above are the short-context tier. Ask where the cut-off is and what longer prompts cost.
Can the workload wait? Batch, with a 24-hour target, and Flex, a preview that may be unavailable at times, each cut the token rate in half.
Do you need PTU yet? PTU is billed every hour and cannot be paused. Deploy first, then reserve.
Which currency and exchange rate will apply to your bill? Prices are set in US dollars. AED and SAR figures here use the pegs, not market rates. Microsoft's page prices are estimates, so confirm your rate in the Azure pricing calculator or under your Enterprise Agreement.