AI inference recommendation engine
Composite and per-model prices pulled directly from provider APIs and rebuilt every hour.
Intelligence verified against the AA index, then divided into price so rankings reward value, not cheapness.
Whether a provider keeps your tokens, for how long, and under whose jurisdiction. Matched on provider statements today.
What could go wrong using this inference, stated in terms that can be checked. Not yet scored.
Who attests each claim today, and who verifies it
The Standard Inference Token (SIT) tracks the cost of producing one million GPT-4-Turbo-equivalent inference tokens, the commodity unit for AI compute. Basket: cheapest qualifying model per provider, eligibility set relative to the scored-model population (top 40%), based on Artificial Analysis v4.3.
→ Read methodologyQuality-adjusted price across 677 models
Grouped by quality tier, ranked within tier by Cost/IQ: verified price per million tokens per unit of AA Intelligence Index. Input, output, and blended prices per model.
Quality here means intelligence only. Latency and uptime verification is in development; neither is scored on this page.
Subscribed to by the teams that buy inference
Create a free account for unlimited recommendations and API access. The engine on this page is the same API your agents can call.
Free for 1,000 requests/day. No credit card required.