Cheaper Inference
Cheaper Inference is a hosted marketplace operated by Keak AI, Inc. for accessing artificial intelligence (AI) models through one application programming interface (API). Applications send requests to the marketplace, which routes them to eligible providers and deducts usage charges from a prepaid workspace balance. The service supports OpenAI-compatible request formats and an Anthropic-compatible Messages API.[1][2]
Origins and marketplace
According to Fabrice Mayrand, Cheaper Inference grew out of an internal effort to buy unused compute commitments below list price. Keak describes three sources of capacity or discounts for the marketplace: volume agreements with AI labs, price competition between data centers serving open-weight models, and unused commitments or spare capacity purchased from suppliers.[3][4]
Suppliers can apply to sell inference capacity through the marketplace. The public seller page describes a process in which suppliers verify their identity and connect a key from a supported network. Cheaper Inference tests which models the key can serve. Keak also advertises dedicated GPU rentals as a separate offering.[4]
API and request routing
The API catalog identifies the models available to a client and the capabilities of each model. Besides text generation, the documentation describes endpoints for image and video generation, embeddings, and document reranking. Supported input formats and features such as streaming depend on the endpoint and model.[2]
According to the API documentation, a request can move to another eligible provider if the first route runs out of capacity or fails before output begins. This fallback preserves the public model identifier and response format. A different serving route can change the amount charged for the request.[2]
Billing and wallet funding
The service prices text inference in U.S. dollars per million tokens and can price media in other units. The workspace wallet can reserve an estimated charge before a request runs and settle the final amount against actual usage. The API documentation says the final charge follows the successful route and is capped at the model maker's applicable direct list price. The price tracker EasiAPI has documented changes in the marketplace's listed prices between observations.[1][2][5]
The service's documentation describes payment card and cryptocurrency top-ups for the workspace balance. It describes Stripe processing for crypto top-ups and leaves the supported assets and networks unspecified.[2]
Request data and retention
Keak's privacy policy says the selected model provider receives the request payload. Keak may keep a copy of prompts and responses after ordinary requests. The policy describes zero data retention (ZDR) as an account-level setting or a per-request option where supported. Keak says that requests covered by ZDR are routed only to eligible providers and that it does not retain their content. Operational and billing records, such as request status, model, usage and settled charge, are recorded separately.[6]
OmniRoute
Cheaper Inference publishes OmniRoute, an AI gateway created by Diego Souza and released under the MIT License. Users can deploy OmniRoute on their own infrastructure and connect provider accounts. Cheaper Inference also offers a hosted edition and bills inference separately from the hosting subscription.[7][8]
References
- ↑ 1.0 1.1 Terms and Conditions. Cheaper Inference. Retrieved 8 November 2026.
- ↑ 2.0 2.1 2.2 2.3 2.4 Build with Cheaper Inference. Cheaper Inference. Retrieved 8 November 2026.
- ↑ Fabrice Mayrand — activity and launch account. LinkedIn. Retrieved 8 November 2026.
- ↑ 4.0 4.1 Cheaper Inference. Cheaper Inference. Retrieved 8 November 2026.
- ↑ Cheaper Inference pricing and review. EasiAPI. Retrieved 8 November 2026.
- ↑ Privacy Policy. Cheaper Inference. Retrieved 8 November 2026.
- ↑ OmniRoute. GitHub. Retrieved 8 November 2026.
- ↑ OmniRoute by Cheaper Inference. Cheaper Inference. Retrieved 8 November 2026.