At 200 requests/user/day (totaling 30,000,000 queries/mo), running Data extraction & tagging on Llama 3.2 3B Instruct costs $2,175.00 / month ($0.435 / user / mo).
Total expected API invoice for 5,000 Active Users generating 30,000,000 queries.
Direct inference cost per MAU to model your SaaS pricing tiers and gross margins.
Savings generated by exploiting 20% cache hits on prompt context.
| Model | Provider | Monthly Spend (5,000 Active Users) | Cost / User / Mo | Annual Spend | Context Limit |
|---|---|---|---|---|---|
| Llama 3.2 3B Instruct (Current) | meta | $2,175.00 | $0.435 | $26,100.00 | 128,000 |
| GPT-5.6 Luna | openai | $18,840.00 | $3.768 | $226,080.00 | 1,050,000 |
| GPT-5.4 mini | openai | $70,650.00 | $14.13 | $847,800.00 | 256,000 |
| GPT-5.4 nano | openai | $19,215.00 | $3.843 | $230,580.00 | 128,000 |
| GPT-4o mini | openai | $12,600.00 | $2.52 | $151,200.00 | 128,000 |
| Claude Haiku 4.5 | anthropic | $86,700.00 | $17.34 | $1,040,400.00 | 1,000,000 |
At 200 queries per user/day with Llama 3.2 3B Instruct, the estimated cost is $0.435 per monthly active user (MAU). Total monthly bill for 5,000 Active Users is $2,175.00.
Assuming a 20% cache hit rate on repeated system and context tokens, prompt caching saves $0.00 per month ($0.00/year) on Llama 3.2 3B Instruct.
With a direct COGS of $0.435 per user/month on Llama 3.2 3B Instruct, charging at least $2.175 per user/month ensures an 80%+ software gross margin.