If you’ve poked around GitHub Copilot’s billing page and walked away more confused than when you started, you’re not alone. Terms like “premium requests,” “model multipliers,” and “included models” get thrown around without much context. This post breaks it all down clearly, using Copilot Pro as the concrete example.

What Even Is a Request?

Before getting to the “premium” part, let’s nail down what a request means. A request is any single interaction where you ask Copilot to do something: sending a message in chat, triggering an inline suggestion, asking the CLI a question. Each time you hit send, that’s one request.

Not all requests are created equal, though. Some use more compute than others, and that’s where the premium vs. non-premium split comes in.

What Makes a Request “Premium”?

Premium requests are interactions that use advanced models or features beyond Copilot’s baseline. On a paid plan like Copilot Pro, three models are considered “included” and consume ZERO premium requests: GPT-4.1, GPT-4o, and GPT-5 mini. You can chat with those models all day long without touching your premium request allowance.

The moment you switch to a different model, like Claude Sonnet 4.6, Gemini 2.5 Pro, GPT-5, or any other non-included model, you’re spending premium requests. This is true even for a single back-and-forth chat message. Each user prompt you send counts as one premium request, multiplied by the model’s rate.

Beyond chat, other features also draw from the same premium request pool: Copilot CLI interactions using non-default models, code reviews triggered via the coding agent, and Copilot Spaces sessions all count as premium requests.

The Copilot Pro Allowance: 300 Requests Per Month

Copilot Pro costs $10/month and comes with 300 premium requests per month, per user. That allowance resets on the 1st of every month at midnight UTC. Unused requests do not roll over.

300 requests sounds like a lot until you factor in model multipliers.

Model Multipliers: Why One Chat Can Cost More Than One Request

This is where most people get tripped up. Not every premium model costs 1 premium request per message. GitHub uses multipliers based on each model’s complexity:

• Claude Haiku 4.5 costs 0.33x per prompt (about 3 prompts per premium request)
• Claude Sonnet 4.6, Gemini 2.5 Pro, and GPT-5 cost 1x per prompt
• Claude Opus 4.5 and 4.6 cost 3x per prompt (one message eats 3 premium requests)
• Claude Opus 4.1 costs 10x
• Claude Opus 4.6 in fast mode (preview) costs a steep 30x

So if you send 10 messages using Claude Opus 4.5, you’ve consumed 30 of your 300 monthly premium requests in one short conversation. Burn through a longer agentic session with a high-multiplier model and you can drain your allowance surprisingly fast.

One useful tip from the docs: if you’re on a paid plan and use auto model selection in Copilot Chat in VS Code, you get a 10% discount on multipliers. Claude Sonnet 4.6 would cost 0.9x instead of 1x, for example.

What Happens When You Hit the Limit?

If you exhaust your 300 monthly premium requests on Copilot Pro, you don’t lose access to Copilot entirely. You can still use the three included models (GPT-4.1, GPT-4o, GPT-5 mini) for the rest of the month with no additional charge, though response times may vary during high demand.

If you want to keep using premium models beyond your allowance, you need to set up a paid overage budget. Additional premium requests are billed at $0.04 per request. One important catch: accounts created before August 22, 2025 have a default $0 budget, meaning overages are blocked unless you explicitly update that setting in your billing preferences.

The Quick Mental Model

Think of it this way. Your Copilot Pro subscription covers two things:

1. UNLIMITED chat and code suggestions with the included models (GPT-4.1, GPT-4o, GPT-5 mini)
2. A 300 premium request monthly budget for everything else, drained according to each model’s multiplier

If you mostly chat with GPT-4o or GPT-5 mini, you’ll almost never touch your premium request balance. If you regularly reach for Claude Opus or advanced reasoning models, 300 requests can go faster than expected.

Wrapping Up

The core confusion usually comes down to one misunderstanding: people assume every chat message costs a premium request. That’s only true for non-included models. Stick to GPT-4.1, GPT-4o, or GPT-5 mini and your premium balance doesn’t move at all. Switch to a high-multiplier model and the math changes quickly.

Check your current usage at any time in Copilot settings on GitHub or directly in your IDE. If you find yourself burning through your allowance regularly, either adjusting which model you reach for by default or upgrading to Copilot Pro+ (which includes 1500 premium requests/month) are both reasonable options.

If you want to go from beginner to pro with GitHub Copilot, check out my full Udemy course on mastering the tool: