At 200 requests/user/day (totaling 60,000 queries/mo), running Data extraction & tagging on Qwen 2.5 Coder 32B costs $28.68 / month ($2.868 / user / mo).
Total expected API invoice for 10 Active Users generating 60,000 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 (10 Active Users) | Cost / User / Mo | Annual Spend | Context Limit |
|---|---|---|---|---|---|
| Qwen 2.5 Coder 32B (Current) | qwen | $28.68 | $2.868 | $344.16 | 128,000 |
| GPT-5.3 Codex | openai | $382.20 | $38.22 | $4,586.40 | 256,000 |
| DeepSeek Coder V2.5 | deepseek | $17.976 | $1.798 | $215.712 | 128,000 |
| Codestral 2501 | mistral | $43.02 | $4.302 | $516.24 | 256,000 |
| Grok Build 0.1 | xai | $130.80 | $13.08 | $1,569.60 | 256,000 |
| DeepInfra — Qwen 2.5 Coder 32B | deepinfra | $13.20 | $1.32 | $158.40 | 128,000 |
At 200 queries per user/day with Qwen 2.5 Coder 32B, the estimated cost is $2.868 per monthly active user (MAU). Total monthly bill for 10 Active Users is $28.68.
Assuming a 20% cache hit rate on repeated system and context tokens, prompt caching saves $4.32 per month ($51.84/year) on Qwen 2.5 Coder 32B.
With a direct COGS of $2.868 per user/month on Qwen 2.5 Coder 32B, charging at least $14.34 per user/month ensures an 80%+ software gross margin.