or-lookup-models

name	or-lookup-models
description	Use when the user wants to look up models in the OpenRouter catalog — fetch the full model list, look up a specific model by ID, or get raw catalog data for downstream filtering. Triggers on phrases like "look up an OpenRouter model", "get the OpenRouter catalog", "what does OpenRouter list for <model-id>", "fetch OR models". This is the foundation skill the other OR research skills build on.

OpenRouter Model Lookup

Fetch the OpenRouter model catalog and return either the full catalog or a single model's record. The endpoint is unauthenticated — no API key required.

When to use

• User asks for raw catalog data, a specific model's metadata, or a list of models
• Another skill in this plugin needs catalog data to filter or rank models

Endpoint

GET https://openrouter.ai/api/v1/models

Returns { "data": [ { ...model... }, ... ] }. Each model has fields including:

• id — e.g. openai/gpt-4o-mini
• name — display name
• created — unix timestamp
• description
• context_length — max context tokens
• architecture.input_modalities — array, e.g. ["text", "image"]
• architecture.output_modalities — array
• architecture.tokenizer
• pricing.prompt — USD per token (string)
• pricing.completion — USD per token (string)
• pricing.image, pricing.request, pricing.web_search
• top_provider.context_length, top_provider.max_completion_tokens
• supported_parameters — array, e.g. ["tools", "tool_choice", "response_format", "structured_outputs"]
• per_request_limits (nullable)

Workflow

1. Fetch the catalog with curl or WebFetch:

   curl -s https://openrouter.ai/api/v1/models -H "Accept: application/json"
   

2. If the user named a specific model ID, filter to that record and return it cleanly formatted.
3. If they asked for the full catalog, return a brief summary (count, sample of IDs) — never dump the whole JSON to the user unless they ask.
4. If they want a slice (e.g. "all Anthropic models"), filter data[].id by prefix and return matching records.

Pricing notes

Prices are returned as strings in USD per token (not per million). To display per-1M-tokens, multiply by 1,000,000. Always show prices as $X.XX / 1M tokens for readability unless the user wants raw values.

Output

• Single model: a compact summary block showing id, context, modalities, pricing per 1M, supported parameters.
• Multiple models: a markdown table with columns: ID, Context, In Modalities, Prompt $/1M, Completion $/1M.
• Always cite the data source as the OpenRouter catalog endpoint and note the timestamp of fetch.

Caching

The catalog changes frequently. Do not cache results across invocations — re-fetch each time the skill runs.