Live-rate model table
Published list rates, 2026-09-29. Click a column to sort, click a row to load that model into the calculator. Prices are USD per million tokens; ctx is the context window.
| Model ID | Name | In $/M | Out $/M | Ctx (k) |
|---|
Cheap end, expensive end
The 434 priced models span a wide band. Median rates: $0.45 per million input, $2.00 per million output.
Cheapest inputs:
~deepseek/deepseek-v4-flash-latest— $0.012 in, $1.25 outibm-granite/granite-4.0-h-micro— $0.017 in, $0.112 outdeepseek/deepseek-v4-flash-0731— $0.018 in, $0.32 outopenai/gpt-oss-20b— $0.018 in, $0.09 outmistralai/mistral-nemo— $0.019 in, $0.03 out
Priciest outputs:
openai/o1-pro— $600 outopenai/gpt-5.5-pro— $180 outopenai/gpt-5.4-pro— $180 outopenai/gpt-5.2-pro— $168 outopenai/gpt-5-pro— $120 out
How OpenRouter pricing works
OpenRouter forwards your request to the underlying provider and bills the provider's list price against your credit balance. One formula covers every model:
(input tokens / 1M) × input rate + (output tokens / 1M) × output rate
- Model choice is the biggest lever. A 10x input-rate gap between frontier and efficient tiers is normal — routing non-critical traffic to a cheaper variant saves more than any other tweak.
- Check the detail page before committing. Caching, batch endpoints, and reasoning tokens can shift the effective rate.
- Billing is unified. One monthly export from OpenRouter shows per-model spend — no stitching together five provider dashboards.
- Estimate from real traffic. Pull average input and output tokens from your logs, multiply by the rates in the table above, and scale by daily volume. The calculator does this in one step.