At 10 requests/user/day (totaling 30,000 queries/mo), running RAG / search-augmented answers on Qwen 2.5 72B Instruct costs $56.70 / month ($0.567 / user / mo).
RAG / search-augmented answers for a 100 active users (public launch) audience means 1,000 requests per day on average, each carrying 8,000 input and 500 output tokens at a 50% cacheable share — so the effective per-request cost on Qwen 2.5 72B Instruct is $0.00189, of which $37.80/mo comes back from prompt caching. Per active user that's $0.567/mo — the number to hold against your pricing tier and gross margin. Charging at least $2.835/user/mo keeps gross margin at or above 80%.
Quick answer
For 100 Active Users, RAG / search-augmented answers on Qwen 2.5 72B Instruct is modeled at 30,000 requests per month, costing $56.70 per month or $0.567 per active user.
Method & trust
This projection combines the selected active-user tier, requests per user per day, input/output tokens per request, and the modeled cache share. It is an operating estimate, not a provider invoice; production traffic and retries should be measured against it.
Total expected API invoice for 100 Active Users generating 30,000 queries.
Direct inference cost per MAU to model your SaaS pricing tiers and gross margins.
Savings generated by exploiting 50% cache hits on prompt context.
| Model | Provider | Monthly Spend (100 Active Users) | Cost / User / Mo | Annual Spend | Context Limit |
|---|---|---|---|---|---|
| Qwen 2.5 72B Instruct (Current) | qwen | $56.70 | $0.567 | $680.40 | 128,000 |
| GPT-5.6 Terra | openai | $444.00 | $4.44 | $5,328.00 | 1,050,000 |
| GPT-5.4 Workhorse | openai | $555.00 | $5.55 | $6,660.00 | 256,000 |
| o3-mini | openai | $264.00 | $2.64 | $3,168.00 | 200,000 |
| o4-mini | openai | $231.00 | $2.31 | $2,772.00 | 256,000 |
| o1-mini | openai | $264.00 | $2.64 | $3,168.00 | 128,000 |
At 10 queries per user/day with Qwen 2.5 72B Instruct, the estimated cost is $0.567 per monthly active user (MAU). Total monthly bill for 100 Active Users is $56.70.
Assuming a 50% cache hit rate on repeated system and context tokens, prompt caching saves $37.80 per month ($453.60/year) on Qwen 2.5 72B Instruct.
With a direct COGS of $0.567 per user/month on Qwen 2.5 72B Instruct, charging at least $2.835 per user/month ensures an 80%+ software gross margin.