# Cheaper Inference Cheaper Inference (cheaperinference.com) is a discount API gateway for [[Large Language Models (LLMs)]] operated by Keak AI, Inc., a Delaware company. It exposes OpenAI- and Anthropic-compatible endpoints, ranks the eligible provider routes by estimated cost on every request, and starts with the cheapest available. Headline claim: save up to 60%, with market-linked rates never above direct provider list price and no separate routing surcharge. It runs on a prepaid wallet, $5 minimum. What makes it structurally different from [[OpenRouter]] or [[Ramp Router]] is that it is a **meta-aggregator**: its own documentation names OpenRouter and [[Vercel]] AI Gateway as upstream routes. It aggregates aggregators, so a prompt can pass through three custodians before it reaches a model. ## Data handling The privacy policy is better than the category average on the part that usually matters: prompt and response bodies are not stored in its application database, and Keak states it does not use them to train Keak models. Operational metadata is retained — model, endpoint, timestamps, usage, settled charge, key attribution, latency, provider-attempt data — with product and security logs kept around 12 months and billing records up to 7 years. Zero data retention is **opt-in per request** (`data_retention: {"mode": "zero"}`, `zdr: true`, or a header), and switching it on restricts routing to Vercel AI Gateway and OpenRouter routes with server-enforced ZDR. That is a real tension: the ZDR-compliant route set and the cheapest route set are not the same, so the privacy switch works against the reason you came. ## The unexplained part The site never publishes where the discount comes from. The only hint is a seller-side pitch inviting people to "turn unused inference capacity into revenue", which suggests a two-sided capacity marketplace. For open-weight models that is ordinary arbitrage. For closed frontier models like Claude and GPT — both named on the homepage — a sub-list-price route implies resold committed capacity, resold credits, or deliberately sacrificed margin, and none of those is disclosed. That is the general lesson worth keeping: a cost router's savings claim is only as trustworthy as its explanation of where the money comes from. An unexplained discount on a good you cannot legally resell is a question, not a feature. ## Integrations Documented support for [[Claude Code]], [[OpenAI Codex]], [[Cline]], [[Cursor]], [[Aider]], Continue, Open WebUI and LibreChat — the target user is someone running coding agents at volume. ## References - Official website: https://cheaperinference.com/ - Documentation: https://cheaperinference.com/docs - Integrations: https://cheaperinference.com/docs/integrations - Privacy policy: https://cheaperinference.com/privacy - Keak subprocessors: https://platform.keak.com/subprocessors ## Related - [[OpenRouter]] - [[Ramp Router]] - [[Large Language Models (LLMs)]] - [[AI Inference]] - [[Vercel]] - [[Claude Code]] - [[OpenAI Codex]] - [[Cline]] - [[Cursor]] - [[Aider]]