GitHub Copilot Premium Request Multipliers: Current Table

Posted on Oct 21 • Originally published at olumia.dev

A GitHub Copilot premium request multiplier is the number that decides how many premium requests a single model interaction deducts from your monthly allowance. Ask a model with a 1x multiplier one question in Copilot Chat and you spend one premium request. Ask a model with a 27x multiplier the same question and you spend 27. The premium request multiplier is, in other words, the price of model choice, and the current table runs from 0.25x at the bottom to 27x at the top, a 108x spread for the same interaction.

There is one thing to know before anything else: as of June 1, 2026, multipliers only apply to Copilot Pro and Pro+ subscribers on annual plans that stayed on the legacy request-based billing model. Every monthly plan (Pro, Pro+, Business, and Enterprise) has moved to usage-based billing with GitHub AI Credits, where model choice shows up as per-token rates instead. This guide covers both sides: the current multiplier table for those still on it, and how the same "model choice multiplies cost" logic survived the billing change for everyone else.

Under request-based billing, each paid Copilot plan included a monthly allowance of premium requests. Every interaction with a premium feature (a chat message, an agent-mode step, a code review) deducted from that allowance according to the model's multiplier: one interaction on a 6x model consumed six premium requests, one on a 0.33x model consumed a third of one. GitHub set each multiplier based on the model's complexity and the resources it consumes to serve.

The mechanics were simple, but they made the same workload cost wildly different amounts depending on which model a developer happened to pick from the dropdown. If premium requests as a unit are new to you, start with our plain-language explainer on what a premium request is, then come back for the multipliers.

On June 1, 2026, GitHub replaced premium request units with GitHub AI Credits for all monthly plans. Multipliers survived in exactly one place: Copilot Pro and Pro+ subscribers on an existing annual plan, who remain on legacy request-based billing until their plan expires.

For the full picture of how the request-based system worked: per-plan allowances, overage billing, and the timeline of the transition, see our complete reference on GitHub Copilot premium requests, allowances, and billing.

These are the legacy multipliers in effect for annual Pro and Pro+ subscribers, per GitHub's published table. GitHub trims the table as models retire, so it is shorter now than it was in the summer.

Lightweight models: 0.25x to 0.33x. One premium request buys three interactions at 0.33x.

Claude Opus 4.8 is also one of the models GitHub is retiring on November 16, although individual subscribers on annual plans can keep it.

Run the arithmetic once and the stakes are obvious: 100 interactions on Claude Haiku 4.5 consume 33 premium requests. The same 100 interactions on Claude Opus 4.8 consume 2,700. Identical workload, identical allowance: an 82x difference in burn rate, decided entirely by a dropdown.

Copilot code review carries a multiplier of 13, meaning every review of a pull request or in-IDE review deducts 13 premium requests. A team that wires automatic review into every PR is spending 13 premium requests per pull request before any human asks a single chat question, one of the quietest ways an allowance disappears mid-month.

If you use auto model selection in Copilot Chat, Copilot CLI, the GitHub Copilot app, or the Copilot cloud agent, GitHub applies a 10% discount to the multiplier: a 1x model bills at 0.9x. It is a real discount, but note what it is not: a guarantee about which model gets selected, and therefore which base multiplier you start from.

For everyone not on a legacy annual plan, Copilot usage now consumes GitHub AI Credits, where 1 credit equals $0.01. Usage is calculated from token consumption (input, output, and cached tokens) at the published per-token rate for each model. Plan prices did not change: Pro is $10/month with $10 in monthly credits included, Pro+ is $39 with $39, Business is $19 per user with $19, and Enterprise is $39 per user with $39. Business and Enterprise customers ran on promotional allowances of $30 and $70 per user from June 1 to September 1, 2026. That bump has ended and the standard $19 and $39 allowances now apply, a cliff we covered in what the credit cliff changed.

Code completions and next edit suggestions remain unlimited and consume no credits. The old fallback behavior (exhaust your allowance, drop to a lower-cost model, keep working) is gone; usage is now governed by available credits and admin budget controls, with credits poolable across an organization.

Here is the part that matters for anyone tempted to file multipliers under "legacy trivia": under AI credits, model choice multiplies cost just as hard, only per token instead of per request. GitHub's current pricing spans $0.20 per million input tokens for GPT-5.6 Luna to $10.00 for Claude Fable 5, and $1.20 per million output tokens at the bottom against $50.00 at the top.

Work one example: a coding session that consumes 500K input tokens and 100K output tokens costs about $0.22 on GPT-5.6 Luna, and about $10.00 on Claude Fable 5. Same session, roughly a 45x spread. That is a multiplier by any other name, and it compounds across every developer, every day. The discipline of watching which models handle which work, what we call model mix efficiency, did not get less important when the multiplier table became legacy. It became the single biggest lever on the bill.

If you have annual Pro or Pro+ subscribers on the legacy model, make sure whoever owns the budget knows the multipliers increased in June, that the heaviest model on the table now runs 27x, and that automatic code review bills 13x per PR. Then make the conversion decision deliberately, before expiration makes it for you.

If you are on AI credits (which, for organizations, is now everyone), the question shifts from "which multiplier" to "which model mix, at what burn rate, across which teams." That question rarely stops at one vendor: the same engineering organization typically connects several AI coding assistants (GitHub Copilot, Cursor, Claude, and whatever a team adopts next), each metering model usage in its own units at its own rates. Normalizing that into one per-team, per-model view is the groundwork we cover in our guide to tracking AI code assistant spend across every vendor, and it is the layer Olumia builds for engineering leaders: read-only connections to each assistant's admin data, so the model-mix lever is visible before the invoice lands.

A GitHub Copilot premium request multiplier is the per-model exchange rate on your monthly allowance, currently 0.25x to 27x, applicable only to annual Pro and Pro+ plans still on legacy request-based billing, and 13x for every Copilot code review. For everyone else, the June 1 move to GitHub AI Credits replaced the table with per-token rates that spread just as wide. Either way, the lesson is the same one the multiplier table taught from the start: with AI coding assistants, the model your developers pick is the price you pay.