This chapter covers the pricing models, tiers, and cost management strategies for Azure Cognitive Services, a critical topic for the AI-900 exam. Understanding how Cognitive Services are priced — from free tiers to committed tiers — directly impacts solution design and cost optimization. Approximately 10-15% of AI-900 exam questions touch on pricing, tiers, and cost considerations for AI workloads, often asking you to choose the most cost-effective tier for a given scenario or to identify the correct SLA for a specific tier. This chapter will equip you with the precise knowledge needed to answer these questions confidently.
Jump to a section
Imagine a cafeteria that serves AI-powered meals. The cafeteria offers three service tiers: Free, Standard (Pay-as-you-go), and Enterprise (Committed). The Free tier is like a sample counter where you get a small cup of soup each day — you can taste it, but you can't serve a crowd. The soup is prepared using basic ingredients, and you get only one cup per day. The Standard tier is like a regular meal plan: you pay per item you take. Each bowl of soup costs a fixed price, but if you take many bowls, the cost adds up. There's no reservation; you just walk in and pay. The Enterprise tier is like a catering contract: you commit to ordering a certain number of meals each month (e.g., 1,000 bowls) and pay a lower per-bowl price. Even if you don't eat all 1,000 bowls, you still pay for the commitment. The cafeteria also has a loyalty card (free tier) that gives you a limited number of free bowls each month, but you can't choose the soup flavor — it's whatever is pre-selected. The Standard plan lets you choose any flavor and pay per bowl. The Enterprise plan gives you a dedicated chef and priority service, but you must sign a contract. This mirrors Azure Cognitive Services pricing: Free tier offers limited transactions per month with restricted features (e.g., only 5,000 text translations per month, no SLA). Standard (S0) tier charges per transaction with a pay-as-you-go model, includes an SLA of 99.9% uptime, and supports all features. Enterprise (Committed) tier requires a minimum commitment (e.g., $1,000/month) for discounted per-unit rates and reserved capacity, but you pay even if you under-utilize.
What Are Cognitive Services Pricing Tiers?
Azure Cognitive Services are billed based on the pricing tier you select for each resource. The tier determines the number of transactions allowed per time period, the features available, the service-level agreement (SLA), and the cost per transaction. Microsoft offers three main categories of tiers: Free (F0), Standard (S0), and sometimes additional tiers like S1, S2, etc., depending on the service. Some services also offer a 'Committed Tier' (previously called 'Reserved Capacity') for high-volume workloads.
Why Pricing Tiers Exist
Pricing tiers allow Microsoft to offer Cognitive Services at different price points to accommodate various usage patterns. The free tier encourages experimentation and learning. The standard tier is for production workloads with predictable scaling. Committed tiers provide cost savings for customers with steady, high-volume usage. Each tier is designed to match a specific customer profile:
Free (F0): Ideal for proof-of-concepts, personal projects, and learning. Limits are low (e.g., 5,000 transactions per month for Translator, 20 transactions per minute for Computer Vision). No SLA.
Standard (S0): For production applications. Pay per transaction. Includes a 99.9% SLA for most services. Higher rate limits (e.g., 1,000 transactions per minute).
Committed Tier: For high-volume, predictable workloads. You commit to a minimum amount of transactions per month (e.g., 1 million transactions) and receive a discounted price per transaction. You pay for the committed amount even if you use less.
How Pricing Works Internally
When you create a Cognitive Services resource in Azure, you select a pricing tier. This tier is enforced at the API level. Each API call includes a key that identifies the resource and its tier. Azure's infrastructure meters each call, applying rate limits and counting transactions against the tier's allowance. For the free tier, once the monthly limit is reached, the API returns a 403 Forbidden error (or a 429 Too Many Requests if rate-limited). For standard tiers, you are billed per transaction based on the published price.
Key Components, Values, and Defaults
Free Tier (F0): Typically offers 5,000-20,000 transactions per month, depending on the service. For example, Text Analytics free tier allows 5,000 text records per month. Computer Vision free tier allows 20 transactions per minute. No SLA. You can only have one free tier resource per subscription per region.
Standard Tier (S0): Pay-as-you-go. Prices vary by service. For example, Translator text translation costs $10 per million characters. Speech-to-text costs $1 per hour of audio. Most standard tiers have a 99.9% SLA. Rate limits are higher (e.g., 1,000 transactions per minute for many services).
Committed Tier: You commit to a specific volume (e.g., 1 million, 10 million, 100 million transactions per month) and pay a flat fee. The per-unit cost is lower than standard. For example, for Translator, a commitment of 1 million characters per month might cost $8 per million (vs $10 pay-as-you-go). You must commit for a month; if you exceed the committed amount, you pay the standard rate for overages.
SLA: Free tier has no SLA. Standard tier typically offers 99.9% uptime. Some services like Speech Services offer 99.95% for real-time transcription. Committed tiers have the same SLA as standard.
Rate Limits: Free tier has strict rate limits (e.g., 20 calls per minute). Standard tier has higher limits (e.g., 1,000 calls per minute). Committed tiers may have even higher limits.
Configuration and Verification Commands
You can create a Cognitive Services resource with a specific tier using Azure CLI:
az cognitiveservices account create --name myResource --resource-group myGroup --kind TextAnalytics --sku F0 --location westus --yesTo change the tier after creation:
az cognitiveservices account update --name myResource --resource-group myGroup --sku S0To view the current tier:
az cognitiveservices account show --name myResource --resource-group myGroup --query properties.skuHow Pricing Interacts with Other Azure Services
Azure Monitor: You can monitor usage and costs using Azure Monitor metrics. Set alerts when approaching free tier limits.
Cost Management: Use Azure Cost Management to analyze spending across Cognitive Services resources.
Azure Policy: Enforce specific tiers across subscriptions using Azure Policy.
Azure Reservations: For committed tiers, you can purchase reservations for 1 or 3 years for additional savings (up to 40% off pay-as-you-go).
Common Exam Scenarios
The AI-900 exam often presents a scenario where you need to select the appropriate pricing tier. For example:
A startup building a prototype with limited budget should use the Free tier.
A production application expecting 100,000 API calls per month should use Standard tier.
A company with a steady 10 million calls per month should consider Committed tier for cost savings.
If an SLA is required, you cannot use the Free tier.
Edge Cases and Exceptions
Some services offer multiple standard tiers (e.g., S0, S1, S2) with different rate limits and prices. For example, Computer Vision has S0, S1, S2 tiers. S0 is for up to 10 transactions per second (TPS), S1 for 30 TPS, S2 for 60 TPS.
The free tier is not available for all services. For example, Personalizer does not have a free tier.
Some services like Anomaly Detector have a free tier that allows up to 20,000 transactions per month.
The committed tier for some services requires a minimum of $1,000 per month.
You can have multiple free tier resources if they are in different regions, but only one per region per subscription.
Summary of Pricing Model
Tier | Transactions | Cost | SLA | Rate Limits
Free (F0) | 5,000/month | Free | None | Low (e.g., 20/min)
Standard (S0)| Per call | Pay-per-call | 99.9% | High (e.g., 1,000/min)
Committed | Min. 1M/month| Flat fee + overage | 99.9% | Very highIdentify Workload Requirements
First, determine the expected number of API calls per month, the need for an SLA, and the budget. For example, a prototype with <5,000 calls/month and no SLA requirement is a candidate for the Free tier. A production app with 50,000 calls/month and 99.9% uptime requirement needs Standard tier. A high-volume call center with 2 million speech-to-text hours per year should evaluate Committed tier.
Create Resource with Selected Tier
Using Azure Portal, CLI, or PowerShell, create a Cognitive Services resource specifying the SKU (e.g., F0, S0). For example, in Azure Portal, during creation, under 'Pricing Tier', select the desired tier. In CLI: `az cognitiveservices account create --sku F0`. The tier is set at creation and can be changed later.
Configure Rate Limits and Quotas
After creation, you can view and adjust rate limits (if applicable) in the 'Limits' blade of the resource. For standard tiers, rate limits are predefined but can be increased by requesting a quota increase. For free tier, limits are fixed. For committed tier, rate limits are typically higher and can be adjusted based on the commitment level.
Monitor Usage and Costs
Use Azure Monitor to track the number of API calls, latency, and error rates. Set up alerts to notify when approaching free tier limits (e.g., at 80% usage). Use Azure Cost Management to view daily costs. For standard tiers, costs are incurred per transaction. For committed tiers, you are billed the flat fee monthly, with overage charged at standard rates.
Scale Up or Change Tier as Needed
If usage grows beyond the free tier limit, upgrade to Standard tier. If usage becomes predictable and high, switch to Committed tier. To change tier, use `az cognitiveservices account update --sku S0`. Note that changing from free to standard may incur costs immediately. Also, downgrading from standard to free is possible only if the resource has not exceeded free limits in the current month.
Optimize Costs with Reservations
For committed tier users with long-term, stable usage, purchase Azure Reservations for Cognitive Services to save up to 40% compared to pay-as-you-go. Reservations are applied at the subscription level and require a 1-year or 3-year commitment. For example, if you commit to $1,000/month for 1 year, you get a discount on the per-unit price.
Scenario 1: E-commerce Product Recommendations
An online retailer wants to use the Personalizer service to recommend products to customers. They expect 500,000 API calls per month. They need an SLA because downtime directly impacts revenue. The solution: Use Standard tier (S0) for Personalizer. The cost is $1.50 per 1,000 calls, so monthly cost is $750. They set up Azure Monitor alerts for error rates and cost thresholds. If traffic doubles, they consider switching to Committed tier, which would reduce per-call cost to $1.20 per 1,000 calls with a $1,000/month minimum commitment. They also purchase a 1-year reservation for the committed tier to save an additional 20%.
Scenario 2: Multilingual Customer Support Chatbot
A global company deploys a chatbot using Translator and Language Understanding (LUIS). The chatbot handles 10 million text translations per month and 5 million LUIS predictions. They need high availability. Using Standard tier for Translator would cost $10 per million characters = $100/month for translation (assuming 10 million characters). LUIS standard tier costs $1.50 per 1,000 calls = $7,500/month. Total $7,600/month. They evaluate Committed tier for LUIS: commit to 5 million calls/month at $1.20 per 1,000 calls = $6,000/month, saving $1,500/month. They also use Azure Reservations for LUIS to save an additional 15%. They monitor usage to ensure they don't exceed the committed amount significantly.
Scenario 3: Healthcare Document Analysis
A hospital uses Text Analytics for Health to extract medical entities from patient records. They process 2 million documents per month. They require HIPAA compliance and a 99.95% SLA. Text Analytics for Health is available only in Standard tier (S0) with a 99.9% SLA. To achieve 99.95%, they deploy multiple instances in paired regions and use Azure Traffic Manager. The cost is $1 per 1,000 text records = $2,000/month. They consider Committed tier: commit to 2 million records/month at $0.85 per 1,000 records = $1,700/month, saving $300/month. They also implement cost allocation tags to track usage by department.
Common Misconfiguration: A startup uses the Free tier for a production app and hits the transaction limit, causing the app to fail. They should have started with Standard tier or at least monitored usage and set alerts. Another mistake: choosing Committed tier for a variable workload; they end up paying for unused capacity. They should use Standard tier with auto-scaling instead.
The AI-900 exam (Objective 1.1: Identify features of common AI workloads) tests your understanding of when to use each pricing tier. Specifically, you may be asked to:
Select the appropriate tier for a given scenario (e.g., a prototype vs. production).
Identify the correct SLA for a tier (Free has none, Standard has 99.9%).
Recognize the limitations of the Free tier (e.g., max 5,000 transactions per month, no SLA, low rate limits).
Understand that committed tiers offer lower per-unit costs but require a minimum commitment.
Common Wrong Answers: 1. Choosing Free tier for a production app because it's free. Wrong because free tier has no SLA and low limits; production apps need reliability and scale. 2. Assuming all tiers have the same SLA. Wrong: Free tier has no SLA; Standard has 99.9%. 3. Thinking you can use Free tier for unlimited time. Wrong: Free tier has monthly limits; once exceeded, API calls fail. 4. Believing committed tier is always cheaper. Wrong: Only cheaper if you use the committed volume; if usage is low, you overpay.
Specific Numbers to Memorize: - Free tier: 5,000-20,000 transactions/month (varies by service). - Standard tier SLA: 99.9%. - Rate limits: Free ~20-100 calls/min; Standard ~1,000 calls/min. - Committed tier: Minimum commitment often $1,000/month. - Reservations: 1-year or 3-year commitment, up to 40% savings.
Edge Cases: - Some services like Speech Services offer a 99.95% SLA for real-time transcription (not just 99.9%). - The free tier for some services (e.g., Translator) allows 2 million characters per month, not 5,000 transactions. - You can have multiple free tier resources if they are in different regions. - Changing from standard to free is only allowed if the resource has not exceeded free limits in the current billing month.
How to Eliminate Wrong Answers: - If the scenario mentions 'prototype', 'testing', or 'learning', the answer is likely Free tier. - If 'production', 'SLA', or 'high volume' is mentioned, eliminate Free tier. - If 'cost-effective for high volume' or 'predictable usage', consider Committed tier. - If 'flexible scaling' or 'variable workload', Standard tier is best. - Always check if SLA is required; if yes, Free is out.
Free tier (F0) is for experimentation with no SLA and low transaction limits (e.g., 5,000/month).
Standard tier (S0) is for production with a 99.9% SLA and pay-per-transaction pricing.
Committed tier offers discounted per-unit prices but requires a minimum monthly commitment (e.g., $1,000).
Only one free tier resource per subscription per region is allowed.
Downgrading from Standard to Free is only possible if usage is within free tier limits for the month.
Azure Reservations can save up to 40% on committed tiers with 1- or 3-year terms.
The AI-900 exam often tests which tier to choose based on scenario requirements (SLA, volume, cost).
Rate limits for free tier are typically 20-100 calls per minute; standard tier allows 1,000+ calls per minute.
These come up on the exam all the time. Here's how to tell them apart.
Free Tier (F0)
Cost: $0 per month
Transactions: 5,000-20,000/month (varies by service)
SLA: None
Rate Limits: Low (e.g., 20 calls/min)
Best for: Prototyping, learning, non-critical apps
Standard Tier (S0)
Cost: Pay per transaction
Transactions: Unlimited (subject to rate limits)
SLA: 99.9%
Rate Limits: High (e.g., 1,000 calls/min)
Best for: Production apps with variable or growing usage
Mistake
Free tier is suitable for production applications because it's free.
Correct
Free tier has no SLA, very low rate limits (e.g., 20 calls/min), and a monthly transaction cap (e.g., 5,000 calls). Production applications require reliability and scalability, so Standard tier is mandatory.
Mistake
All Cognitive Services tiers have the same 99.9% SLA.
Correct
Only Standard (S0) and above tiers have an SLA. Free tier (F0) has no SLA. Some services like Speech Services have a 99.95% SLA for real-time transcription.
Mistake
Committed tier is always cheaper than Standard tier.
Correct
Committed tier offers lower per-unit cost but requires a minimum monthly commitment. If you don't reach the committed volume, you still pay the full commitment, making it more expensive than pay-as-you-go for low usage.
Mistake
You can have unlimited free tier resources per subscription.
Correct
You can only have one free tier resource per subscription per region. However, you can create free tier resources in different regions, each counting as separate.
Mistake
Changing from Standard to Free tier is always allowed.
Correct
You can downgrade to Free tier only if the resource has not exceeded the free tier's monthly transaction limit in the current billing cycle. If it has, the downgrade will be blocked.
Reveal each answer, then mark whether you got it right. Score 60%+ to unlock the next chapter.
Free tier (F0) is free but has low transaction limits (e.g., 5,000 per month) and no SLA. Standard tier (S0) charges per transaction, offers a 99.9% SLA, and has higher rate limits. For production, always use Standard or higher. For learning, Free is sufficient.
Yes, you can change the pricing tier of an existing Cognitive Services resource at any time using the Azure Portal or CLI. The change takes effect immediately with no downtime. However, if you exceed free tier limits during the month, you cannot downgrade back to free until the next billing cycle.
Standard tier (S0) and above typically have a 99.9% uptime SLA. Free tier has no SLA. Some services like Speech Services offer 99.95% for real-time transcription. Check the specific service documentation for exact SLA details.
Use Azure Monitor metrics for the Cognitive Services resource. Set up alerts for 'TotalCalls' metric with a threshold at 80% of the free tier limit. You can also use Azure Cost Management to track costs. For free tier, set a budget alert to notify you when approaching the free limit.
Once the free tier monthly transaction limit is reached, subsequent API calls return a 403 Forbidden error (or 429 if rate-limited). The service stops processing requests until the next billing month. To avoid disruption, upgrade to Standard tier.
No, Committed tier is not available for all services. It is typically offered for high-volume services like Translator, Speech, and Text Analytics. Check the pricing page for each service. For services without Committed tier, use Standard tier with reservations if available.
Yes, you can create one free tier resource per service per subscription per region. For example, you can have one free Translator resource and one free Computer Vision resource in the same subscription. But you cannot have two free Translator resources in the same region.
You've just covered Cognitive Services Pricing and Tiers — now see how well it sticks with free AI-900 practice questions. Full explanations included, no account needed.
Done with this chapter?