Simulating realistic Document summarization parameters (25,000 in / 600 out with 10% cache reuse). DeepInfra — Qwen 2.5 Coder 32B delivers a 58% cost reduction over Qwen3 Coder Flash (OpenRouter).
| Traffic Volume Tier | Qwen3 Coder Flash (OpenRouter) Monthly | DeepInfra — Qwen 2.5 Coder 32B Monthly | Monthly Savings by picking DeepInfra — Qwen 2.5 Coder 32B |
|---|---|---|---|
| 1,000 reqs/mo (Dev/Testing) | $5.07 | $2.144 | Save $2.926 / mo |
| 10,000 reqs/mo (Small App) | $50.70 | $21.44 | Save $29.26 / mo |
| 100,000 reqs/mo (Growth Production) | $507.00 | $214.40 | Save $292.60 / mo |
| 1,000,000 reqs/mo (Scale SaaS) | $5,070.00 | $2,144.00 | Save $2,926.00 / mo |
DeepInfra — Qwen 2.5 Coder 32B is 58% cheaper for Document summarization workloads. At standard Document summarization parameter ratios (25,000 input tokens, 600 output tokens, 10% cache hit), DeepInfra — Qwen 2.5 Coder 32B costs $0.002144 per request compared to $0.00507 on Qwen3 Coder Flash (OpenRouter).
Qwen3 Coder Flash (OpenRouter) offers a context window of 1,000,000 tokens (max output: 65,536), while DeepInfra — Qwen 2.5 Coder 32B offers 128,000 tokens (max output: 8,192).
At 100,000 requests per month, using DeepInfra — Qwen 2.5 Coder 32B saves $292.60 every month (or $3,511.20 annually) compared to Qwen3 Coder Flash (OpenRouter).
Long-context models pay off here — compare price per 1M tokens at your true document size. Summarize once, store the result; don't re-summarize unchanged documents. For batch backfills, nightly jobs can use cache-friendly request ordering.