If you use AI APIs regularly, you know that model costs can add up quickly. Whether you're building an AI agent, shipping a SaaS product, or processing thousands of requests, inference costs directly affect your margins. At CheapestToken, we're taking a different approach to pricing. Our discounts come from two different strategies, depending on how each model is delivered.
Open-Weight Models: We Host, You Save
For open-weight models, we handle the hosting ourselves. Instead of relying entirely on another provider's API pricing, we deploy supported models on our infrastructure and manage the resources required to serve requests. This gives us greater control over deployment and serving costs. We can build pricing around infrastructure expenses and resource utilisation, allowing us to offer selected models at lower rates. For developers, that means access to capable models without paying more than necessary for inference. We host open-weight models in FP16, not quantised versions (yet).
Proprietary Models: Discounts at Our Expense
Proprietary models work differently. Their weights aren't available for independent hosting, so we access them through their respective providers. That means we still have to pay the underlying API costs. So how are we offering discounts? For selected proprietary models, we're absorbing part of the cost ourselves to provide lower introductory rates. Essentially, we're subsidising your usage. These offers are temporary. We cannot absorb the difference indefinitely, and selected models may return to their regular pricing once the promotional period ends. If you're already planning to test or integrate these models, now is a good time to take advantage of the current rates.
One API. One Wallet. Pay Per Token.
You don't need separate billing arrangements for every model. Add credits to your CheapestToken wallet, generate an API key, and start making requests through our OpenAI-compatible API. Your balance is deducted according to the tokens consumed and the selected model's pricing. No subscription commitment. No need to manage multiple model integrations just to experiment with different options.
Check the Current Pricing
We want developers to spend more time building and less time worrying about inference bills. Explore the available models, compare input and output rates, and find the right fit for your workload. Our open-weight model pricing reflects our hosting approach. Our proprietary model discounts are limited-time offers and may change as upstream costs evolve.