Translations

AI Cost Estimate

As of May 26, 2026

1. Models and Pricing

OpenAI Models

Model	Usage Type	Price
GPT-5 mini	Input tokens	$0.25 per 1M tokens
GPT-5 mini	Cached input tokens	$0.025 per 1M tokens
GPT-5 mini	Output tokens	$2.00 per 1M tokens
text-embedding-3-small	Embeddings	$0.020 per 1M tokens

OpenAI lists GPT-5 mini at $0.25 per 1M input tokens and $2.00 per 1M output tokens, and text-embedding-3-small at $0.02 per 1M tokens. (OpenAI Developers) (OpenAI Developers)

Agat AI Classification Model Benchmark

For security classification, this estimate uses an Agat AI classification model priced similarly to Claude Haiku 4.5:

Model Benchmark	Usage Type	Price
Claude Haiku 4.5 equivalent	Input tokens	$1.00 per 1M tokens
Claude Haiku 4.5 equivalent	Output tokens	$5.00 per 1M tokens

Anthropic lists Claude Haiku 4.5 at $1 per million input tokens and $5 per million output tokens. It also notes that Claude models are available on Amazon Bedrock, with partner cloud pricing and endpoint behavior handled by the cloud provider. (Claude Platform)

Note: If Agat AI classification is benchmarked against Claude Haiku 3.5 on Bedrock instead of Claude Haiku 4.5, the rate is lower: $0.80 per 1M input tokens and $4.00 per 1M output tokens. Claude Haiku 3.5 is marked as retired except on Bedrock and Vertex AI. (Claude Platform)

2. Averages and Assumptions

Document Size Assumption

An average document, such as a PDF or Word file, is assumed to contain:


3.5 pages × 500 tokens per page = 1,750 tokens per document

Standard Private AI User Activity


Average questions per user per month: 25 questions
Average document uploads per user per month: 20 documents

BGD15 Query — Questions


SELECT AVG(qu.numOfQuestions)
FROM (
  SELECT COUNT(*) AS numOfQuestions 
  FROM BG_USERS u
  JOIN BG_QUESTIONS q ON u.InternalUserId = q.CreatedBy
  WHERE u.LastUsedTime IS NOT NULL
  GROUP BY u.InternalUserId, MONTH(q.CreatedTime)
) AS qu;

BGD15 Query — Documents


SELECT AVG(cu.numOfFiles)
FROM (
  SELECT COUNT(*) AS numOfFiles 
  FROM BG_USERS u
  JOIN BG_CONTENT_LIST c ON u.InternalUserId = c.CreatedBy
  WHERE u.LastUsedTime IS NOT NULL
    AND c.ContentType = 'Document'
  GROUP BY u.InternalUserId, MONTH(c.CreatedTime)
) AS cu;

Token Consumption per Question


Average input tokens per question: 10,000
Average output tokens per question: 200

BGD15 Query — Token Usage


SELECT 
  AVG([RequestPromptTokenCount]), 
  AVG([RequestCompletionTokenCount])
FROM [AI_USAGE_HISTORY]
WHERE [RequestType] = 'Query' 
  AND RequestSourceType = 'Chat' 
  AND RequestSourceId IS NOT NULL;

3. Function Definitions


InputCostLLM(llmModel, numOfTokens)
= Cost of input tokens for the given LLM model.

OutputCostLLM(llmModel, numOfTokens)
= Cost of output tokens for the given LLM model.

CostEmbedding(embeddingModel, numOfTokens)
= Cost of embedding tokens for the given embedding model.

ClassificationInputCost(classificationModel, numOfTokens)
= Cost of input tokens for the classification model.

ClassificationOutputCost(classificationModel, numOfTokens)
= Cost of output tokens for the classification model.

4. Product Parameters


LLM Model: GPT-5 mini
Embedding Model: text-embedding-3-small
Security Classification Model: Agat AI classification model, priced like Claude Haiku 4.5

5. Cost Formulas

GPT-5 mini Input Cost


InputCostLLM(GPT-5 mini, numOfTokens)
= ($0.25 / 1,000,000) × numOfTokens
= 2.5 × 10^-7 × numOfTokens

GPT-5 mini Output Cost


OutputCostLLM(GPT-5 mini, numOfTokens)
= ($2.00 / 1,000,000) × numOfTokens
= 2 × 10^-6 × numOfTokens

text-embedding-3-small Embedding Cost


CostEmbedding(text-embedding-3-small, numOfTokens)
= ($0.020 / 1,000,000) × numOfTokens
= 2 × 10^-8 × numOfTokens

Agat AI Classification Input Cost

Using Claude Haiku 4.5-equivalent pricing:


ClassificationInputCost(numOfTokens)
= ($1.00 / 1,000,000) × numOfTokens
= 1 × 10^-6 × numOfTokens

Agat AI Classification Output Cost

Using Claude Haiku 4.5-equivalent pricing:


ClassificationOutputCost(numOfTokens)
= ($5.00 / 1,000,000) × numOfTokens
= 5 × 10^-6 × numOfTokens

6. Cost Examples

Average GPT-5 mini Cost per Question

Assumption:


Input tokens per question: 10,000
Output tokens per question: 200

Calculation:


Input cost:
10,000 × 2.5 × 10^-7 = $0.0025

Output cost:
200 × 2 × 10^-6 = $0.0004

Total cost per question:
$0.0025 + $0.0004 = $0.0029

Result


Average GPT-5 mini cost per question: $0.0029
Approx. $1.00 per 345 questions

Average Embedding Cost per Document

Assumption:


Average document size: 1,750 tokens

Calculation:


1,750 × 2 × 10^-8 = $0.000035

Result


Average embedding cost per document: $0.000035
Approx. $1.00 per 28,571 documents

7. Private AI Monthly Cost per User

Questions Cost


25 questions/month × $0.0029 per question
= $0.0725 per user/month

Document Embedding Cost

Using the original average of 20 documents per user per month:


20 documents/month × $0.000035 per document
= $0.0007 per user/month

Total Private AI Monthly Cost per User


$0.0725 + $0.0007 = $0.0732 per user/month

Result


Private AI cost per average user: $0.0732 per month

8. Private AI Example Scenarios

Number of Users	Monthly Cost
100 users	$7.32/month
1,000 users	$73.20/month
10,000 users	$732/month
100,000 users	$7,320/month

9. Security Users — Document and Prompt Classification

Security users require additional classification because Agat AI classifies:


1. Documents uploaded by users
2. Prompts sent to external AI providers such as ChatGPT, Claude, Gemini, etc.

Security User Assumptions


Document uploads per security user per day: 3 documents
Days per month: 30
Documents per security user per month: 3 × 30 = 90 documents

Average prompts/questions per user per month: 25 prompts

Average document size: 1,750 input tokens
Average prompt/request payload: 10,000 input tokens
Average classification output: 50 output tokens

The 10,000-token prompt classification assumption uses the same average input-token estimate from the existing AI usage query. If the security classifier only scans the raw user prompt and not the full enriched request payload, the actual classification cost will be lower.

Security Classification Cost per Document

Using Claude Haiku 4.5-equivalent pricing:


Input:
1,750 tokens × $1.00 / 1,000,000
= $0.00175

Output:
50 tokens × $5.00 / 1,000,000
= $0.00025

Total classification cost per document:
$0.00175 + $0.00025 = $0.0020

Result


Average security classification cost per document: $0.0020
Approx. $1.00 per 500 classified documents

Security Classification Cost per Prompt


Input:
10,000 tokens × $1.00 / 1,000,000
= $0.0100

Output:
50 tokens × $5.00 / 1,000,000
= $0.00025

Total classification cost per prompt:
$0.0100 + $0.00025 = $0.01025

Result


Average security classification cost per prompt: $0.01025
Approx. $1.00 per 98 classified prompts

10. Security Classification Monthly Cost per User

Document Classification


90 documents/month × $0.0020
= $0.18 per security user/month

Prompt Classification


25 prompts/month × $0.01025
= $0.25625 per security user/month

Total Security Classification Cost


$0.18 + $0.25625 = $0.43625 per security user/month

Result


Security classification cost per user: $0.43625 per month

11. Full Monthly Cost per Security User

If a security user also uses the Private AI flow, and we use the higher upload volume of 90 documents/month, the full cost is:

GPT-5 mini Questions


25 questions/month × $0.0029
= $0.0725

Embeddings for 90 Uploaded Documents


90 documents/month × $0.000035
= $0.00315

Security Classification


$0.43625

Total Full Security User Cost


$0.0725 + $0.00315 + $0.43625
= $0.51190 per security user/month

Result


Full cost per security user: $0.51190 per month

12. Security User Example Scenarios

Classification Only

Number of Security Users	Monthly Classification Cost
100 users	$43.63/month
1,000 users	$436.25/month
10,000 users	$4,362.50/month
100,000 users	$43,625/month

Full Security User Cost

Includes GPT-5 mini questions, document embeddings, and security classification.

Number of Security Users	Monthly Full Cost
100 users	$51.19/month
1,000 users	$511.90/month
10,000 users	$5,119/month
100,000 users	$51,190/month

13. Summary of Unit Costs

Item	Approx. Unit Cost
GPT-5 mini question	$0.0029 per question
Document embedding	$0.000035 per document
Security document classification	$0.0020 per document
Security prompt classification	$0.01025 per prompt
Private AI average user	$0.0732 per user/month
Security classification only	$0.43625 per user/month
Full security user	$0.51190 per user/month

14. Optional Lower-Cost Classification Variant

If the Agat AI classifier is priced like Claude Haiku 3.5 rather than Claude Haiku 4.5, the security classification cost is approximately 20% lower:


Claude Haiku 3.5-equivalent pricing:
Input: $0.80 per 1M tokens
Output: $4.00 per 1M tokens

Using that rate:


Document classification:
$0.0016 per document

Prompt classification:
$0.0082 per prompt

Monthly security classification:
90 documents × $0.0016 + 25 prompts × $0.0082
= $0.144 + $0.205
= $0.349 per security user/month

So the full security user cost would be:


$0.0725 GPT-5 mini questions
+ $0.00315 embeddings
+ $0.349 classification
= $0.42465 per security user/month

Add this section at the end:

Recap


Private AI average user:
$0.0732 per user/month

Security user cost:
$0.51190 per security user/month

Main cost drivers:


Workspace question using GPT-5 mini 
$0.0029 per question

Document embedding using OpenAI embeddings
$0.000035 per document

Model Pricing Comparison

GPT-4o mini

Context size: 128K tokens

Token type	Price
Input tokens	$0.15 per 1M tokens
Output tokens	$0.60 per 1M tokens
Reasoning tokens	Not separately billed / included in output pricing

Claude Haiku 4.5 with Reasoning

Context size: 1M tokens

Token type	Price
Input tokens	$1.00 per 1M tokens
Output tokens	$5.00 per 1M tokens
Reasoning tokens	Billed as output tokens: $5.00 per 1M reasoning tokens

Claude Sonnet 4.6 with Reasoning

Context size: 1M tokens

Token type	Price
Input tokens	$3.00 per 1M tokens
Output tokens	$15.00 per 1M tokens
Reasoning tokens	Billed as output tokens: $15.00 per 1M reasoning tokens

Workspace Question Cost with Classification

Model	Approximate cost per prompt
GPT-4o mini	~$0.003
Claude Haiku 4.5 with reasoning	~$0.20
Claude Sonnet 4.6 with reasoning	Higher than Haiku, depending on token usage

Summary

GPT-4o mini is significantly cheaper for this workflow.

Claude Haiku 4.5 is more expensive, especially when reasoning tokens are used, because reasoning tokens are billed as output tokens.

Claude Sonnet 4.6 is the most expensive option in this comparison, but it may provide stronger reasoning and coding quality depending on the task.

For scale:

Users	Private AI Monthly Cost	Full Security Monthly Cost
100	$7.32	$51.19
1,000	$73.20	$511.90
10,000	$732	$5,119
100,000	$7,320	$51,190

Model Pricing Comparison

GPT-4o mini

Context size: 128K tokens

Token type	Price
Input tokens	$0.15 per 1M tokens
Output tokens	$0.60 per 1M tokens
Reasoning tokens	Not separately billed / included in output pricing

Claude Haiku 4.5 with Reasoning

Context size: 1M tokens

Token type	Price
Input tokens	$1.00 per 1M tokens
Output tokens	$5.00 per 1M tokens
Reasoning tokens	Billed as output tokens: $5.00 per 1M reasoning tokens

Claude Sonnet 4.6 with Reasoning

Context size: 1M tokens

Token type	Price
Input tokens	$3.00 per 1M tokens
Output tokens	$15.00 per 1M tokens
Reasoning tokens	Billed as output tokens: $15.00 per 1M reasoning tokens

Workspace Question Cost with Classification

Model	Approximate cost per prompt
GPT-4o mini	~$0.003
Claude Haiku 4.5 with reasoning	~$0.20
Claude Sonnet 4.6 with reasoning	~$0.70