Pick what you're building, tweak a few numbers, and compare every major model side by side. Agent loops, caching, batch and long-context pricing included. No token math.
| Model | Quality | $ / 1M in · out · cached | Monthly | Per user / mo | Per 1k req |
|---|
★ cheapest · Quality = Artificial Analysis Intelligence Index (higher is better; "—" = no score in data). Badges: tier/batch published rates, est. fallback estimate, cache-write write cost counted.
Pro tier coming soon: saved setups, price-drop alerts, CSV export. One email when it launches, no spam.
Requests per month = active users × messages per user per day × 30. Each request is one LLM call (or, in agent mode, one call per step). Cost per call = input tokens × input price + output tokens × output price, using each model's published per-million-token prices from OpenRouter.
Agents re-send their whole context every step. Step n includes the system prompt, RAG context, user input, and everything earlier steps produced (tool results, output, reasoning). Costs grow roughly with the square of the step count, which is why coding agents are expensive.
The stable prefix (system prompt, plus earlier agent steps) can be cached. The hit-rate share is billed at the cached-read price. When a model publishes a cache-write price (e.g. Anthropic, Gemini), the missed share is billed at that write price instead of the normal input price.
Where OpenRouter lists a :batch variant we use its exact rates for the batched share of traffic; otherwise we assume 50% off and mark it "est.". Long-context tiers (higher prices above e.g. 200k or 272k input tokens) use each model's published tiers; models with none get a flat ×2 input / ×1.5 output estimate if the toggle is on.
Most traffic is easy. Route it to the cheapest model that fits your context (and accepts images, if you use them) and send the hard share to the highest-quality model you selected. You can override either pick; it resets when you switch templates.
Image or audio output, web search fees, retries, provider minimums, free tiers, router/classifier overhead, hosting and taxes. Prices are OpenRouter list prices and can differ slightly from buying direct.