Save money on AI API costs: get model tokens at discounted prices

AI features have a habit of starting cheap and getting expensive. A prototype that cost a few dollars a day becomes a production service billing thousands a month, and the growth is rarely where you expect it. Before looking for a lower price, it's worth understanding what you're actually paying for, because that tells you where a discount will matter most.

Where an AI API bill comes from

Almost every large language model API bills per token, with separate prices for the tokens you send and the tokens the model writes back. Prices are quoted per million tokens, and output is typically several times more expensive than input. A few other meters show up on modern models too.

• Input tokens: your system prompt, the conversation history, any documents or tool results you include, and the user's message.
• Output tokens: everything the model generates, including reasoning tokens on models that think before answering.
• Cached input: repeated context that the provider has cached, billed at a fraction of the normal input price.
• Cache writes: the one-off cost of putting context into the cache so later requests can read it cheaply.

The surprise for most teams is input volume. Chat applications resend the whole conversation on every turn, agents stuff tool outputs back into the context, and retrieval pipelines attach long passages to each question. A single user session can easily send the same few thousand tokens dozens of times. Output is pricier per token, but input is usually where the volume lives.

Why the same model can cost less

When you call a model through its creator's API, you pay the published list price. That isn't the only way to reach the same model. Capacity is sold through several channels, and an API gateway can buy it on better terms than an individual developer, then resell it per token.

A gateway like GreenTokens sits between your code and the upstream providers. You send a normal OpenAI- or Anthropic-format request; the gateway authenticates it, forwards it to a provider that serves the model, streams the response back and bills you for the tokens used. Because the gateway buys capacity in aggregate and can route each model to a provider that serves it for less, it can charge less than the list price while you still get the same model, the same output format and the same token counts.

What a discount is worth in practice

Percentages are easy to quote and hard to feel, so here's a worked example with round, hypothetical numbers. Imagine a model with an official price of $3 per million input tokens and $15 per million output tokens, and an application that uses 500 million input tokens and 50 million output tokens a month.

At a 60% discount the same workload costs $1,350 less every month, or more than $16,000 a year, without touching a line of product code. Real discounts vary by model, so check current prices rather than assuming a flat rate. On GreenTokens, the models page lists our price per million tokens next to the official list price for every model and every token type, so you can run this calculation with real numbers.

What to check before you switch

A lower price is only a saving if nothing else gets worse. Before moving production traffic to any discounted provider, work through these questions.

1. Is it really the same model? Look for the provider's own model IDs and compare responses on a few of your real prompts.
2. Can you see the official price next to theirs? A provider confident in its discount will show you both, per model and per token type.
3. How is usage billed? The fair answer is on the exact token usage the upstream model reports, with no rounding up to the nearest thousand and no charge for requests you cancel.
4. Does it work with your existing code? OpenAI- and Anthropic-compatible endpoints mean you change a base URL and a key, not your integration.
5. What happens when something breaks? Look for a public status page, request IDs you can quote to support, and a clear error format.
6. Can you control spend? Per-key limits and the ability to pause or revoke a key protect you from runaway loops and leaked credentials.
7. What does it cost to start? A subscription or minimum commitment eats into the saving; prepaid, pay-as-you-go billing doesn't.

How GreenTokens approaches it

GreenTokens was built around those questions. The API speaks both the OpenAI and Anthropic formats, so the official SDKs and tools such as Claude Code, Codex CLI and opencode work by changing the base URL and key. Text requests are billed on the exact token usage the provider reports, and cached input is billed at the lower cached rate.

There's no subscription. You top up a prepaid wallet with crypto and spend it per token across every model. Each API key can carry its own daily and monthly spend limit, keys can be paused or revoked instantly, and every request appears in your dashboard with its model, tokens, cost and latency. Per-model availability and 90 days of uptime are published on the status page.

Getting started

Switching takes a few minutes. Sign in with your email, add funds, create an API key, and point your SDK at the GreenTokens base URL. Start by sending a slice of your traffic, compare the responses and the bill, and move the rest once you're satisfied. The cheapest token is still the one you don't send, so pair a lower price with the usage habits in our guide to reducing LLM token costs.