What Are AI Credits? Pricing AI Usage Without Losing Money

AI credits are a pricing unit, not a model feature. They turn the cost of tokens, models and caching into one number a customer can see and budget. GitHub prices Copilot usage in AI credits worth $0.01 each, with cost driven by the model and the input, output and cached token mix (GitHub Docs). That answers the question most AI pricing pages dodge: how to charge when unit cost moves.

TL;DR

• An AI credit is a prepaid unit of model usage, priced so a customer can budget in one number.
• GitHub values one Copilot AI credit at $0.01, and the cost per request depends on the model and token mix.
• Paid Copilot plans include base credits plus a variable flex allotment, so the seller can adapt as AI economics change.
• Unused credits expire at reset, but code completions stay unlimited on paid plans.

Disclosure: ZeroRaised publishes this analysis and is a launch site for bootstrapped products. Facts come from the linked sources as checked on 30 Oct 2026; ZeroRaised has no commercial relationship with the companies named.

AI credit is a prepaid unit of model usage that turns token counts and model prices into a single number a customer can budget against.

What is an AI credit, and why use one?#

A credit is a translation layer. Your cost is measured in tokens that vary by model, input versus output, and whether the prompt hit a cache. Your customer does not want to reason about any of that, so a credit lets you publish one unit, set a conversion rate, and keep the model mix as your problem.

The catch: sell a fixed-price credit and route the request to a more expensive model, and you absorb the difference. Credits work best when you control routing.

How does GitHub price Copilot credits?#

GitHub measures usage in GitHub AI Credits, where one credit equals $0.01, and cost depends on the model and the input, output and cached tokens (GitHub Docs). The plans bundle credits rather than selling them one at a time:

• Pro at $10 per month includes 1,000 base credits plus 500 flex credits, for 1,500 total (GitHub Docs).
• Pro+ at $39 includes 3,900 base plus 3,100 flex, for 7,000 (GitHub Docs).
• Max at $100 includes 10,000 base plus 10,000 flex, for 20,000 (GitHub Docs).

The base number matches the subscription price, because a credit is a cent. The flex number is the interesting part.

Why use base credits plus a flex allotment?#

GitHub describes base credits as matching the subscription price and the flex allotment as a variable allowance designed to adapt as AI economics change (GitHub Docs). That buys room to reprice without a public fight.

If your price list is "1,500 credits for $10", a cost change forces you to cut the allowance or raise the price. If it is "1,000 base plus a flex allotment", the flex number is a variable you can shrink in a bad cost year while the base stays stable. GitHub also offers a temporary 30% discount on Copilot Fusion model costs through 30 Nov 2026, and 10% off when using auto model selection (GitHub Docs).

Auto model selection is a routing decision, and routing protects margin; the LLM routing guide explains the pattern, and the guide to reducing LLM costs covers the mechanics.

What should stay outside the credit meter?#

GitHub keeps code completions and next edit suggestions unbilled, unlimited on paid plans (GitHub Docs). Metering the feature a developer touches every few minutes would make the product feel stingy at the moment it proves its value.

Granola made a similar call, launching dictation on Mac on all plans without limits, months after Wispr launched a meeting notetaker (TechCrunch). When every competitor bundles a feature, charging for it separately is a gift to them. The rule: meter the work that scales with usage, not the habit that makes the product useful.

What happens when credits run out?#

GitHub gives three paths: upgrade and pay the price difference, set a budget in US dollars for extra usage, or wait for the reset (GitHub Docs). The reset is the first of each month at 00:00 UTC, and unused credits do not carry over (GitHub Docs).

For enterprise, Copilot is included for every licensed user, metered from the first request against an enterprise budget with no included allowance (GitHub changelog). Completions and next edit suggestions need no budget; chat, code review and cloud features do (GitHub changelog). The line to copy: "An exhausted budget never removes access" (GitHub changelog).

If a customer hits zero at 2am, does the product stop, degrade, or keep working and bill later? For Enterprise, GitHub's answer is that access stays. Decide yours before launch.

How does routing change the economics of credits?#

Copilot Fusion routes and orchestrates models per task, with Cascade and Critique workflows, where one model drafts and another accepts, escalates or reviews (GitHub changelog). Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of Claude Code with Opus 5.5 Max, at about half the API cost per task (GitHub changelog). It is billed per model used at standard rates, with a 30% discount on paid plans for a limited time (GitHub changelog).

A fixed credit price bleeds money if every request hits the most expensive model.

How would you design credits for your own product?#

Start with the unit. Pick a value a customer recognises, the way one Copilot credit maps to one cent, then decide what it buys and which model paths are included.

Add a reset rule and a cap: monthly expiry keeps the liability bounded, and a dollar budget for overage is the safety valve. Keep one core action unlimited on paid plans, and meter the expensive, variable work.

The AI pricing guide, SaaS pricing guide, AI spend caps analysis and Anthropic credits guide go deeper.

What this means for bootstrapped founders#

• Sell one unit, not a token table. Customers compare credits; they do not compare per-million-token prices across three model tiers.
• Split base from flex. It gives you room to adjust when model costs move without changing the headline price.
• Cap the downside. Offer a dollar budget for overage, and decide what happens at zero before a customer discovers it for you.
• Leave the habit unmetered. Meter the expensive, variable work. Keep the action that proves value free on paid plans.
• Control routing to protect margin. The model you choose is the cost you pay. Credits only work if you own that decision.

Where ZeroRaised fits#

ZeroRaised is a launch and discovery site for products built without institutional venture capital or accelerator equity. If you have a product people can try, submit your product; the pricing page explains the current options and there are no promises of traffic or results.

For more practical analysis, join the ZeroRaised newsletter. It is a plain email, not a growth hack.

Sources#

• GitHub Docs: usage-based billing for individuals
• GitHub changelog: Copilot included with GitHub Enterprise
• GitHub changelog: route and orchestrate models with Copilot Fusion
• TechCrunch: Granola launches dictation