GitHub Copilot's New Multiplier Table: Opus 27x, Sonnet 9x, Codex 6x for Annual Subscribers on June 1

As of June 1, 2026, annual Copilot Pro and Pro+ subscribers staying on request-based billing see their included premium requests worth considerably less. GitHub’s updated multiplier docs¹ list Claude Opus 4.6 and 4.7 at 27x, Sonnet 4.6 at 9x, GPT-5.3 Codex at 6x, and Gemini 3.1 Pro at 6x, effective the moment GitHub’s new AI Credits model takes over. These subscribers chose annual billing specifically to avoid this kind of drift. On May 28, 2026, Anthropic released Opus 4.8⁵ at identical $5/$25 per-million-token pricing to Opus 4.7, so any Copilot multiplier GitHub assigns to Opus 4.8 is expected to land at the same 27x position. Since then, Anthropic added a tier above Opus: Claude Fable 5⁶, released June 9, 2026, was briefly Anthropic’s most capable widely released model, priced at $10/$50 per million tokens on the direct API. Fable 5 on GitHub Copilot uses usage-based billing at provider list pricing — it does not participate in the legacy multiplier system at all. [Updated June 2026] As of June 12, Fable 5 has been disabled globally by Anthropic following a US government export-control directive.

The Multiplier Table: What Annual Subscribers Actually Pay After June 1

GitHub’s docs¹ publish the before-and-after figures for annual plans staying on request-based billing:

Opus 4.7’s multiplier rose in stages: it launched at 7.5x promotional pricing (April 16–30), moved to 15x on May 1 when the promo expired, and reaches 27x for annual request-based subscribers on June 1. The 15x shown above is that immediate pre-June-1 rate.

Opus 4.8 ships at the same $5/$25 per-million-token pricing as Opus 4.7⁵, so it carries no new cost at the direct API level. [Updated June 2026] GitHub’s multiplier docs have since confirmed Opus 4.8 sits at 27x on the legacy annual billing system, the same as Opus 4.7. The quality case for Opus 4.8 is material: Anthropic reports it is four times less likely to allow flaws in code versus Opus 4.7, with a SWE-Bench Pro score of 69.2% against Opus 4.7’s 64.3%. For teams using direct API access outside Copilot, swapping Opus 4.7 for Opus 4.8 costs nothing. For teams inside Copilot’s multiplier system, the confirmed 27x position means the billing math is identical to Opus 4.7.

On June 9, 2026, Anthropic released Claude Fable 5⁶, a new Mythos-class tier positioned above Opus 4.8. It is Anthropic’s most capable widely released model, priced at $10/$50 per million tokens on the direct API, exactly twice Opus 4.8’s rate. GitHub made Fable 5 available to Copilot Pro+, Max, Business, and Enterprise users — notably under usage-based billing at provider list pricing, not through the legacy multiplier system. This means annual plan holders on legacy request-based billing are not billed for Fable 5 via a multiplier at all; it requires GitHub’s newer token-metered AI Credits model. [Updated June 2026] That access window was short-lived: on June 12, Anthropic disabled Fable 5 and Mythos 5 for all customers worldwide after a US Commerce Department directive citing export-control authority barred access to foreign nationals — a restriction Anthropic could not enforce selectively in real time. Fable 5 is currently unavailable across all GitHub Copilot experiences and on the direct Anthropic API. For the full account, see US Export Order Forces Anthropic to Disable Fable 5 and Mythos 5 Worldwide.

The arithmetic is blunt. At 27x, a model invocation that consumed one premium request in January now consumes 27. Sonnet moves from parity to 9x. A subscriber who structured their workflow around Sonnet being effectively included at parity now has nine months of assumptions invalidated by a docs update.

The Full Current Table: More Models, Higher Ceiling [Updated June 2026]

Since the June 1 multipliers published, GitHub has continued expanding the legacy annual-plan multiplier table with newly released models. As of mid-June 2026, the docs list a wider range than the six-model table at launch:

Two things stand out in this expanded table. First, GPT-5.5 at 57x is more than double the 27x ceiling that made Opus contentious — any annual subscriber who routes a workflow through GPT-5.5 under legacy billing is consuming requests at a rate that makes the Opus backlash look modest. Second, the 0.33x multiplier on lighter models (Haiku 4.5, GPT-4o, Codex Mini) means those models effectively triple subscribers’ effective capacity, a fact absent from most migration discussions. The practical implication: annual plan holders who can tolerate Haiku 4.5 or GPT-4o quality get three requests for the price of one, while those who need frontier-class capability face multipliers that compress a nominal monthly allowance into a handful of real interactions.

Why Annual Plans Became a Trap: No Rollover, No Grandfathering

The multiplier shift would sting on a month-to-month plan. On an annual plan it creates a structural problem: there is no clean exit that costs nothing.

GitHub’s billing docs² confirm that unused premium requests reset on the 1st of each month at 00:00 UTC, with no carry-over under either the current or new billing system. Whatever is left on May 31 resets. The argument that subscribers could bank requests before the cutover does not hold.

Annual subscribers locked in what they understood to be a fixed rate. GitHub is not raising the per-seat annual price. What changed is the exchange rate between a premium request and a model invocation. The multiplier table is the price hike; it is priced in a unit most subscribers weren’t tracking when they paid.

The Three Exit Options GitHub Offers (And What Each Costs)

GitHub’s docs¹ describe three paths for annual subscribers staying on request-based billing:

1. Stay on the annual plan with the new multipliers until the plan expires, then auto-downgrade to Copilot Free. No immediate out-of-pocket cost, but effective capacity drops June 1.

2. Cancel for a prorated refund. The cleanest exit for subscribers who find the new multipliers unworkable, but it terminates a plan purchased specifically for price stability.

3. Upgrade to a monthly paid plan with prorated credits for the remaining annual value. This moves the subscriber onto the AI Credits model and removes the multiplier problem, but requires mid-cycle billing surgery on a plan whose value proposition was not having to do that.

None of these amounts to “keep what you paid for.”

Community Reaction: From ‘Rug Pull’ to Cursor Migration

GitHub Community Discussion #192948³, the main announcement thread, shows 848 downvotes against 20 upvotes. The top comments don’t debate whether frontier models deserve a premium price. They name alternatives³: Cursor at $20/month flat, Windsurf at $15/month flat (now rebranded to Devin Desktop at $20/month), Claude Code, Cline, and Roo Code.

Discussion #192963⁴ uses sharper language. Annual subscribers describe the change as a “bait-and-switch” and a “rug pull,” specifically noting that annual billing was chosen to lock in a rate. The objection isn’t that Opus is expensive; it’s that the cost of invoking Opus was not part of the advertised annual commitment at the time of purchase.

That reading of the contract is coherent. Annual plans are sold on price certainty. A multiplier table that retroactively reclassifies which models count as effectively included is a change to the plan’s value, even if the per-seat headline price holds.

Competitive Math: Cursor, Windsurf, and Claude Code at Flat Rate

The migration arguments in the community threads treat pricing legibility as the decision criterion. Users cite³ Cursor at $20/month and Windsurf at $15/month as alternatives that don’t require tracking per-model request multipliers. [Updated June 2026] Windsurf has since rebranded to Devin Desktop (June 2, 2026) and raised its Pro plan to $20/month, putting it at price parity with Cursor Pro.

The comparison isn’t fully clean. Cursor and Windsurf have their own usage policies, and the community discussions acknowledge this without letting it override the legibility argument. What these platforms offer is a single number per month, not a request budget that must be mentally cross-multiplied by per-model rates before it means anything. That legibility gap is also closing within GitHub’s own ecosystem: the AI Credits model that monthly subscribers already operate under trades multipliers for direct token metering, which at least makes the cost of a frontier-model call visible as a dollar amount rather than a dimensionless integer. For teams considering what that transition costs, Fable 5 Credit Cliff: What the June 23 Billing Shift Means for Teams works through the credit arithmetic on a per-model basis.

For teams running Opus in automated loops, the June 1 change is a cost increase of roughly an order of magnitude. At 27x¹, a workflow consuming 10 premium requests a month now consumes 270.

The structural pattern is not unique to GitHub. Annual contracts produce a captive subscriber base. When underlying costs shift, multiplier tables let a vendor reprice effective service without touching headline fees. Annual subscribers who locked in for price certainty are learning that “price” was defined more narrowly than they assumed.

For the full history of how the Opus multiplier escalated from 7.5x to 15x to 27x in 60 days, see GitHub Copilot’s Opus 4.7 Multiplier: 7.5x to 15x to 27x in 60 Days. For how the shift from premium request units to token-metered AI Credits changes the billing model entirely, see GitHub Copilot Replaces Premium Request Units With Token-Metered AI Credits on June 1.

Frequently Asked Questions

Does the June 1 multiplier change affect monthly Copilot subscribers the same way?

Monthly subscribers already migrated to the AI Credits model earlier in 2026 and are not subject to the request-based multiplier table at all. The 27x/9x/6x multipliers apply exclusively to annual Pro and Pro+ subscribers who remain on the legacy request-based billing system. Monthly-plan pricing is denominated in credits, not multipliers.

What happens to annual subscribers who do nothing before their plan expires?

They auto-downgrade to Copilot Free at the end of their current billing cycle, losing access to premium models entirely. Copilot Free includes only the base models with a limited monthly request cap and no access to Opus, Sonnet 4.6, or GPT-5.3 Codex at any multiplier.

How does the prorated refund for cancellation actually work?

GitHub calculates the refund based on the remaining full months left in the annual term, subtracting any premium requests already consumed. Subscribers who cancel mid-cycle lose access immediately upon cancellation, there is no grace period to finish a project or export settings.

Are enterprise or team Copilot plans subject to these same multipliers?

The published multiplier table and the three exit options apply only to Individual annual Pro and Pro+ plans. Enterprise and team plans negotiate model access through volume licensing agreements and are billed under a separate structure that does not use the per-request multiplier mechanism described in the Individual-plan docs.

Could GitHub adjust these multiplier values again before the annual plans expire?

GitHub’s docs present the multiplier table as the values effective June 1 for the duration of the legacy annual term, but the documentation does not include a contractual guarantee that the values will remain fixed. The precedent of mid-cycle multiplier changes on annual plans is exactly what triggered the current backlash, so a second adjustment during the same term cannot be ruled out by anything in the published policy.