AI Catalog and Custom Rates

accurate ai cost reporting depends on two things knowing exactly how each model is priced applying the prices you actually pay, not just public list prices amberflo’s ai catalog solves both what the ai catalog is the ai catalog is a continuously updated catalog of public pricing for thousands of ai models across 100+ providers it includes all major providers, including openai anthropic google aws bedrock azure ai etc the catalog is used automatically by amberflo’s rating engine to convert raw usage data into cost no setup is required to get started why this matters most organizations assume ai cost accuracy ends at token counting it does not different providers price the same model differently use different units of measure charge separately for input and output apply negotiated discounts that never show up in public pricing without a centralized pricing catalog and custom rates, your ai cost numbers are wrong what you can see in the ai catalog each catalog entry includes the full pricing context required for accurate rating model identity vendor where the model is accessed from for example, a model accessed via openai directly versus via aws bedrock or azure ai provider the company that owns or publishes the model like openai for gpt 5 model name including version or family mode such as chat, embeddings, image generation, audio, or other specialized modes billing units unit of charge tokens, seconds, images, pages, or other provider specific units input vs output differentiation separate pricing for input usage and output usage when applicable this is critical input and output costs often differ significantly and drive very different optimization decisions zero setup cost accuracy by default by default amberflo automatically applies public list pricing from the ai catalog all ai gateway usage is rated immediately cost appears in dashboards without any configuration this makes it possible to go from raw usage to accurate ai cost in minutes custom contracted rates public pricing is rarely what large organizations actually pay if you have negotiated pricing with a provider, you can define custom contracted rates for any model in the catalog what you can override for each model and provider combination, you can set custom input price custom output price custom unit price for non token models overrides are applied only to your account and do not affect the public catalog how custom rates are applied once a contracted rate is configured all usage for that model is rated using your contracted price input and output costs are calculated separately aggregated cost data reflects your real spend, not list prices no changes are required in your applications or ai gateway configuration multi provider model comparison the ai catalog makes it easy to compare pricing across providers for the same model for example the same model may be available directly from openai the same model may also be offered through aws bedrock or azure ai pricing, units, and discount structures may differ across providers this enables provider level cost comparisons routing decisions based on real cost accurate chargeback across teams using different providers who this is for developers understand why two calls to the same model cost different amounts see how input and output usage affects cost validate pricing when switching providers finops and finance teams ensure ai costs reflect negotiated contracts eliminate pricing assumptions and spreadsheets enable accurate chargeback and showback platform and ai infrastructure teams standardize pricing across the organization enforce cost accuracy at the source support multi provider ai strategies without manual reconciliation key takeaway accurate ai cost management starts with accurate pricing amberflo’s ai catalog centralizes ai model pricing across providers applies pricing automatically with zero setup supports contracted rates for real world enterprise pricing enables trustworthy ai cost analytics and chargebacks this is the foundation for making ai spend measurable, explainable, and optimizable