AI Cost Simulation

1. Overview

This document provides a practical way to estimate monthly AI usage cost for Centrify 360 AI.

AI usage requires a separate Centrify 360 AI add-on license.
The figures below cover model usage only and do not replace the commercial AI add-on license.

All figures in this document are illustrative only.
Actual pricing varies by provider, region, contract, model version, prompt caching policy, and future pricing changes.
Always validate against your provider’s official pricing before making commercial commitments.

2. Baseline Assumptions

This page uses a baseline of 500 queries per month.

For estimation purposes, assume the following per AI request:

  • Average input tokens per request: 90K

  • Cached portion of input: 70%

  • Non-cached portion of input: 30%

  • Average output tokens per request: 1K

This means a typical request behaves approximately like:

  • 63K cached input tokens

  • 27K non-cached input tokens

  • 1K output tokens

At 500 queries per month, that becomes:

  • 31.5M cached input tokens / month

  • 13.5M non-cached input tokens / month

  • 0.5M output tokens / month

These assumptions reflect the fact that a large portion of the semantic-model and instruction context is repeated across requests.

3. Why Caching Changes the Economics

Without caching, every request would be billed as a full 90K-token input interaction.
With caching, repeated context can often be reused, which materially lowers cost.

In many real-world usage patterns, this can reduce effective AI cost by around 70% to 80%.

4. Recommended Models for Evaluation

The following models are reasonable starting points for evaluation:

  • GPT-5.4

  • Claude Sonnet 5

  • DeepSeek V4 Pro

  • Gemini 3.6 Flash

5. Baseline Monthly Cost Comparison (500 Queries)

The table below applies publicly listed pricing to the baseline assumptions above.

Model Cached input rate Standard input rate Output rate Illustrative monthly cost at 500 queries

GPT-5.4

$0.25 / 1M

$2.50 / 1M

$15.00 / 1M

≈ $49.13 / month

Claude Sonnet 5

$0.30 / 1M

$3.00 / 1M

$15.00 / 1M

≈ $64.95 / month

DeepSeek V4 Pro

$0.003625 / 1M

$0.435 / 1M

$0.87 / 1M

≈ $6.85 / month

Gemini 3.6 Flash

$0.15 / 1M

$1.50 / 1M

$7.50 / 1M

≈ $29.48 / month

The monthly figures above are based on the formula:

(31.5M × cached input rate) + (13.5M × standard input rate) + (0.5M × output rate)

All rates are expressed per 1 million tokens.

6. Pricing Sources Used

The estimates above were aligned to public pricing pages available online and checked in July 2026:

  • GPT-5.4: public model catalog pricing showing $2.50 / 1M input, $0.25 / 1M cached input, and $15 / 1M output.

  • Claude Sonnet 5: Anthropic pricing page showing $3 / 1M input, $0.30 / 1M cache hits, and $15 / 1M output.

  • DeepSeek V4 Pro: DeepSeek pricing page showing $0.435 / 1M cache miss, $0.003625 / 1M cache hit, and $0.87 / 1M output.

  • Gemini 3.6 Flash: Google Gemini pricing page showing $1.50 / 1M input, $0.15 / 1M context caching, and $7.50 / 1M output.

Because vendors update pricing frequently, treat this page as a planning aid rather than a contractual price sheet.

7. Monthly Query Simulation

The table below shows how the same token assumptions scale with monthly usage.

Monthly Queries Average Input Tokens per Query Cached Input Tokens Non-Cached Input Tokens Output Tokens

500

90K

31.5M

13.5M

0.5M

1,000

90K

63.0M

27.0M

1.0M

2,500

90K

157.5M

67.5M

2.5M

5,000

90K

315.0M

135.0M

5.0M

10,000

90K

630.0M

270.0M

10.0M

8. Practical Recommendations

  • Start with the 500-query baseline and adjust once you have real usage telemetry.

  • Prefer models and providers with strong prompt caching support.

  • Compare the recommended premium-quality and cost-efficient options against your own workload.

  • Revisit pricing regularly because model pricing changes frequently.

  • Include the separate Centrify 360 AI add-on license in all commercial planning.

9. Related Documentation