Inference Indexer

High inference costs and opaque model pricing stall production deployments. Finding the right balance between intelligence and spend usually involves manual spreadsheet work across dozens of provider pages.

This skill connects to the InferenceIndexer API to automate model selection and price auditing. It queries a verified index of over 70 providers and 690 models to find the most cost-effective endpoint for any specific task. Instead of guessing which model is best, it uses deterministic ranking based on Cost/IQ (quality-adjusted cost) and returns ready-to-use endpoint configurations.

What it does

• Rank models by efficiency using the Cost/IQ metric to find the highest intelligence for the lowest price.
• Audit agent spend with a built-in script that calculates monthly and annual savings by switching providers.
• Filter by compliance including Zero-Data-Retention (ZDR) and EU-sovereign hosting requirements.
• Generate endpoint configs providing the exact base_url and model_id needed to swap providers immediately.
• Track price trends to identify model degradation, pricing changes, or new high-value launches.

How it works

1. Call recommend_models with task constraints like budget, context window, or modality.
2. Review the why string and Cost/IQ ranking to select a verified provider.
3. Run the what_am_i_costing.py script to benchmark current usage against the market.
4. Use the endpoint_config object to update your LLM client pointing to the cheaper host.

Frameworks & tools

This skill requires the inferenceindexer MCP server. It interacts with the InferenceIndexer API and includes a Python 3 utility script for cost auditing.

Why this beats prompting it yourself

LLMs have stale knowledge of pricing and often hallucinate provider capabilities or ZDR status. This skill uses a live, deterministic API index where every result carries verifiable evidence and fresh pricing data (updated every 6 hours).

Use cases

• Selecting the cheapest reliable reasoning model for a complex sub-agent.
• Generating a cost-savings report for a production app spending over $1,000/month on inference.
• Finding EU-sovereign endpoints for a GDPR-compliant data extraction task.
• Comparing intelligence scores (AA Index) across new model releases to justify an upgrade.

Known limitations

Pricing data is refreshed every 6 hours and intelligence scores every 7 days. ZDR and EU-sovereign status are provider-reported and not independently verified by the index.