# AI API Cost Estimator

An unauthenticated API that estimates LLM inference cost for a described
workload, over live pricing for 300+ models (sourced from OpenRouter, refreshed
hourly). No API key required.

## Endpoints

GET /api/models
  List the model catalog with current per-1M and per-token prices.
  Optional: ?q=<substring> to filter (e.g. ?q=claude).

GET /api/estimate
  Estimate cost via query params. Repeat ?model= to compare several.
  Params:
    model            model id or substring (repeatable, or comma-separated)
    tokens_in        input tokens per call        (default 1500)
    tokens_out       output tokens per call       (default 400)
    calls            number of calls per period   (default 1000)
    period           day | week | month           (default day)
    mode             live | batch                 (default live)
    batch_discount   multiplier for batch mode    (default 0.5)
    cache_hit_ratio  0..1 fraction of cached input (default 0)
    reasoning_tokens reasoning tokens per call    (default 0)
  Example:
    /api/estimate?model=anthropic/claude-sonnet-5&model=openai/gpt-5.4&tokens_in=1500&tokens_out=400&calls=50000&period=day

POST /api/estimate
  Same, via JSON body: { "models": ["..."], "tokens_in": 1500, ... }

## Response shape

{
  "workload": { ...echoed inputs... },
  "pricing_fetched_at": "ISO timestamp",
  "cheapest": "model-id",
  "max_monthly_savings": 8100.0,
  "summary": "plain-English one-liner you can relay to a user",
  "results": [ { "modelId", "modelName", "breakdown", "projected": {day,week,month,year}, "notes" } ]
}

All costs are USD. See /api/openapi.json for the full schema.