OpenRouter Model & Pricing Tracker

Tracks OpenRouter's complete model catalog and live per-token pricing — every model it routes to, across every provider — over plain HTTP. No login involved: this is OpenRouter's own public catalog API, the same one the OpenRouter website and every SDK read from before spending an API key on a model.

Scope: catalog, not inference

This Actor reads the model list and its pricing. It does not run inference — calling a model requires your own OpenRouter API key and is a fundamentally different kind of tool (an execution client, not a data-extraction Actor). What this gives you is the market map: which models exist, what they cost, and how that changes over time.

Pricing you can actually compare

OpenRouter publishes price as a per-token decimal string ("0.00000022") — correct for billing, unusable for comparison. This Actor converts it to USD per million tokens, the unit the industry actually quotes in, while keeping the original string so nothing is lost to float rounding at that scale:

"pricing": {
  "prompt_per_million": 0.22, "completion_per_million": 0.66,
  "prompt_per_token_raw": "0.00000022", "completion_per_token_raw": "0.00000066",
  "web_search_per_call": null, "cache_read_per_million": 0.007,
  "cache_write_per_million": null, "is_free": false
}

web_search_per_call is kept separate and unconverted — it is billed per call, not per token, and folding it into the per-million-token conversion would produce a meaningless number.

What you get per model

Provider, context length, input/output modalities, supported API parameters (tools, reasoning_effort, response_format, and so on), when it was added to the catalog, and full pricing. 421 models at last count, spanning every major lab plus long-tail providers.

Input

Field	Type	Notes
category	string	Server-side OpenRouter category filter, e.g. programming.
providers	array	Keep only these providers, e.g. anthropic, openai, deepseek.
maxPricePerMillion	number	Ceiling on prompt price, USD per million tokens.
freeOnly	boolean	Keep only zero-priced models.
minContextLength	integer	Minimum input token capacity.

Suggested uses

• Price-change monitoring. Run daily with incremental-style diffing on your side (compare pricing.prompt_per_million run over run) to catch when a provider cuts or raises prices.
• New-model alerts. Filter by created_at to see what launched since your last run.
• Cost-aware model selection. Use maxPricePerMillion and minContextLength together to shortlist candidates for a specific budget and context need before touching the OpenRouter API itself.

Known limits

• Catalog only, not availability or latency. A model listed here may still have provider-specific rate limits or regional restrictions that only surface at inference time.
• is_free reflects list price, not rate limits. Free-tier models often carry stricter rate limits than OpenRouter's pricing field alone conveys.

Local development

pip install -r requirements.txt
python test_local.py --out sample_output.json
python test_local.py --free-only
python test_local.py --providers anthropic openai --max-price 5

sample_output.json in this folder is real output from a live run, kept so the schema can be reviewed without running anything.