LimitWatch

Home › Guides

How AI billing works: tokens, subscriptions, rate limits and credits

A plain-language guide to the main ways AI companies charge for and limit usage. Every quote comes from an official page and shows the date we last checked it. We give no numbers of our own; for current prices and limits, follow the official links. Examples link to dated changelog entries.

1. Usage-based (token) pricing

APIs usually charge for what you use, counted in tokens (pieces of text), with separate prices for input and output and for cached input.

Tokens used for built-in tools are billed at the chosen model's per-token rates.

Source: OpenAI API pricing, last checked 2026-10-22.

For all other models, cached input tokens do not count toward rate limits and are billed at the cache read rate

Source: Anthropic rate limits, last checked 2026-10-23.

On-Demand Inference Pricing: You are billed in 5-minute windows for the duration your model copy is active starting from the first successful invocation.

Source: Amazon Bedrock pricing (custom model import), last checked 2026-10-22.

Examples of token-price changes:

2. Subscriptions and weekly caps

Consumer and coding-tool plans charge a flat monthly fee and limit usage over time windows instead of per token.

Paid plans have two types of usage limits: a session limit that resets every five hours, and weekly limits that reset once every week.

Source: Claude help center, last checked 2026-10-22.

In addition, to manage capacity and ensure fair access to all users, we may limit your usage in other ways, such as weekly and monthly caps or model and feature usage, at our discretion.

Source: Claude help center: Max plan, last checked 2026-10-22.

Details for Claude: How Claude's weekly usage limits work.

Examples:

3. Rate limits and tiers

Rate limits cap how much you can send per minute or per day. Many APIs raise them as you move up usage tiers.

Rate limits use metrics such as RPM (requests per minute), RPD (requests per day), TPM (tokens per minute), TPD (tokens per day), IPM (images per minute), and audio minutes per minute for some streaming audio models.

Source: OpenAI rate limits, last checked 2026-10-22.

Rate limits are defined at the organization level and at the project level, not user level.

Source: OpenAI rate limits, last checked 2026-10-22.

The rate limits for the Messages API are measured in requests per minute (RPM), input tokens per minute (ITPM), and output tokens per minute (OTPM) for each model class.

Source: Anthropic rate limits, last checked 2026-10-23.

Limits are defined by usage tier

Source: Anthropic rate limits, last checked 2026-10-23.

Your AWS account has default quotas, formerly referred to as limits, for Amazon Bedrock.

Source: Amazon Bedrock quotas, last checked 2026-10-22.

Examples:

4. Credits

"Credits" means different things: prepaid API balance, overage credits on a subscription, or credits included with a plan.

Your organization’s usage tier upgrades automatically as its total credit purchases reach each threshold.

Source: OpenAI rate limits (usage tiers), last checked 2026-10-22.

With usage credits enabled, once one of your included limits is fully used (your five-hour session limit or one of your weekly limits), your requests switch to pay-as-you-go pricing on usage credits and keep working.

Source: Claude help center: usage credits, last checked 2026-10-22.

Monthly API credits included with Max plans can’t be used in Claude Code or the Claude apps, including for extra usage.

Source: Claude help center: Claude Code with Pro or Max, last checked 2026-10-22.

Examples:

5. Aggregators and routers

An aggregator sells access to many companies' models through one API and one bill, and decides which provider serves each request.

OpenRouter routes requests to the best available providers for your model.

Source: OpenRouter docs: provider selection, last checked 2026-10-22.

For each model in your request, OpenRouter’s default behavior is to load balance requests across providers, prioritizing price.

Source: OpenRouter docs: provider selection, last checked 2026-10-22.

Whether to allow backup providers when the primary is unavailable.

Source: OpenRouter docs: provider selection (allow_fallbacks), last checked 2026-10-22.

So the price and limits you get through an aggregator can depend on which underlying provider served the request; see the OpenRouter page for its sources. Self-hosted chat apps are different: they are software, not a paid service.

LibreChat is a self-hosted AI chat platform that unifies all major AI providers in a single, privacy-focused interface

Source: LibreChat repository (README), last checked 2026-10-22.

MIT license

Source: LibreChat repository (license), last checked 2026-10-22.

LibreChat is free, open-source (MIT) software you run yourself. Its repository lists no price and no usage limits of its own; what you pay and the limits you hit come from the provider APIs you connect (prices: not stated by LibreChat, since it does not sell model access).

What this guide does not cover

Browse by company: Amazon Bedrock (AWS) · Anthropic (Claude) · Cursor (Anysphere) · DeepSeek · Fireworks AI · GitHub Copilot · Google (Gemini) · Hugging Face · Ollama · OpenAI (ChatGPT) · Perplexity · Together AI · v0 (Vercel) · Vercel AI Gateway · Windsurf · xAI (Grok). Something wrong? Dispute it.