Claude Opus 4.7’s premium-request multiplier on GitHub Copilot has two documented rates: 7.5x at launch on April 16, 2026, and 15x from May 1. The launch changelog labeled the first “promotional pricing until April 30th.” An editor’s note appended to the same page supplied the second: “Promotional pricing ended on April 30th, and the premium request multiplier was updated to 15x.” The 27x figure that circulated afterward is real but attaches elsewhere: GitHub’s legacy annual-plan table lists 27x against Claude Opus 4.8, and since June 1, 2026, multipliers have applied only to Pro and Pro+ subscribers who stayed on annual plans (GitHub Docs). If you are on a monthly plan, none of these numbers describe your bill anymore.
| Date | Rate | What it covered |
|---|---|---|
| April 16, 2026 | 7.5x | Promotional, “until April 30th” (launch post) |
| May 1, 2026 | 15x | Standing rate, disclosed in a same-page editor’s note |
| June 1, 2026 | Multipliers go legacy | Annual-plan Pro and Pro+ only; monthly plans move to token-metered AI credits (billing announcement, docs) |
| October 19, 2026 | 27x, Opus 4.8 | As of this date, the legacy annual-plan table’s Opus entry; no Opus 4.7 row appears (docs) |
Two rates in 15 days, and 46 days after launch the meter itself was replaced. That sequence, not any single multiplier, is what a budget has to survive.
What the Multiplier Does to a Request Budget
Under the legacy request-based system, a premium-request multiplier scales how much of a plan’s monthly allowance one request consumes. At 7.5x, a single Opus 4.7 request drew 7.5 premium requests; at 15x it drew 15 (changelog). Allowances reset on the first of each month at 00:00 UTC, and unused requests do not carry over (Requests in GitHub Copilot (legacy)). If you exceed the allowance and overages are enabled, extra usage is billed at the standard rate (legacy billing overview).
Two counting rules decide what an agentic loop actually costs, and both are easy to miss. Only the prompts you send count as premium requests; tool calls Copilot makes autonomously to finish a task do not (Requests in GitHub Copilot (legacy)), so a long autonomous run is billed per prompt rather than per action. Auto model selection in Copilot Chat, Copilot CLI, the Copilot app, or cloud agent earns a 10% discount on the multiplier, and Copilot code review carries a flat multiplier of 13 (docs). Both of those figures live on the legacy annual-plan page and stay there: on the AI-credit model that replaced premium requests, code review is billed in credits for its tokens plus GitHub Actions minutes, and the multiplier concept does not apply (billing announcement, models and pricing).
The arithmetic consequence of May 1 is blunt: after the promotional window, a fixed allowance covered half as many Opus 4.7 turns. Cost models written between April 16 and April 30 are off by a factor of two. The multiplier was also always request-level accounting inside Copilot, not a per-token price, so any comparison between a multiplier and an API rate has to be built deliberately from both sides.
The Disclosure Gap: Expiry Published, Successor Rate Not
The launch post disclosed in halves. It named the expiry, April 30th, but not the rate that would follow; the 15x figure became visible in an editor’s note dated May 1, the day after promotional pricing ended. Nothing was concealed. The post-promo rate was published, in writing, on the same page. But an appended note is not where most engineering teams expect to learn that a model wired into agent loops and editor workflows has doubled in cost.
The users most exposed were the ones the changelog was pitching. GitHub’s copy credits Opus 4.7 with “stronger multi-step task performance and more reliable agentic execution” (launch post), exactly the workload that generates the most requests per task and absorbs a per-request doubling hardest. That capability claim comes from GitHub’s early testing, not an independent evaluation.
Users filled the gap with rumor, then with disputes. A community thread titled “Opus 4.7 now 15x?” records one subscriber cancelling 30 Enterprise accounts, with 10 more promised by the end of May, and asserting the multiplier was heading “to 27 or 30 by the end of the month” (discussion #194470). That is a user report of a coming increase, not a published rate. A separate April thread describes a Pro+ user’s usage log showing three models they say they never selected, each billed at the 7.5x rate, which they calculated as roughly an $18 overcharge (discussion #192911). One unverified report, but it is the kind of discrepancy only a manual read of your own usage records will catch.
Where Opus 4.7 Ships, and What It Replaces
Opus 4.7 shipped to Copilot Pro+, Business, and Enterprise, with Business and Enterprise administrators required to enable the model’s policy before it appeared, gradually across the standard surfaces: VS Code, Visual Studio, Copilot CLI, GitHub Copilot Cloud Agent, github.com, GitHub Mobile, JetBrains, Xcode, and Eclipse (changelog). The same post announced the consolidation: “Over the coming weeks, Opus 4.7 will replace Opus 4.5 and Opus 4.6 in the model picker for Copilot Pro+” (changelog).
Four days later the ground moved again. On April 20, GitHub removed Opus models from Copilot Pro entirely, leaving Opus 4.7 as the only Opus on Pro+, paused new signups for Pro, Pro+, and Student, and opened a refund path through May 20 for users the changes did not suit (changelog). On May 28, Claude Opus 4.8 arrived at a 15x multiplier “until Usage Based Billing launches on June 1, 2026” (changelog). The Opus slot in the picker turned over twice in six weeks. A dependency on “the Opus in the picker” is a dependency on whatever GitHub last put there.
Annual-plan subscribers face the sharpest version of this. The legacy multipliers page carries a note that “Users on legacy annual Copilot plans will not receive access to new models and features” (docs). Keeping multiplier billing means being frozen on the model list you had while the picker moves on without you.
June 1 Split the Billing in Two
On April 27, GitHub announced that every Copilot plan would move to usage-based billing on June 1, 2026: premium request units replaced by GitHub AI Credits, metered on input, output, and cached tokens at each model’s listed API rates (billing announcement). Base plan prices stayed put. Two sentences in that announcement deserved more attention than they got. First, “Fallback experiences will no longer be available”: the automatic step-down to a cheaper model when you exhaust your allowance is gone, replaced by credits and admin budget controls. Second, annual Pro and Pro+ subscribers would keep premium-request pricing only until plan expiry, with multipliers increasing on June 1 for them alone, and the plan ending in Copilot Free at expiration unless converted, with prorated credits offered for the remainder (billing announcement). Business and Enterprise seats moved to credits with promotional included usage of $30 and $70 per user per month for June through August (billing announcement).
The documentation now states the supersession plainly and dates it. As of October 19, 2026, both reference pages sit under request-based-billing-legacy, and the multipliers page opens by saying “On June 1, 2026, GitHub moved to usage-based billing,” that its multipliers apply “only to Copilot Pro and Copilot Pro+ subscribers on an existing annual plan who remained on the legacy premium request-based billing model after June 1, 2026,” and that multipliers “do not apply to GitHub’s new usage-based billing model” (docs).
The legacy table, as of October 19, 2026, lists Claude Opus 4.8 at 27x, Claude Sonnet 4.6 at 9x, Claude Haiku 4.5 at 0.33x, GPT-5.5 at 57x, and Copilot code review at 13 (docs). On an annual plan, one Opus 4.8 request costs three Sonnet 4.6 requests. The table lists no Opus 4.7 row, so the legacy page offers no multiplier to quote for Opus 4.7 itself.
That is where the 27x story lands: the rate is real, and the model is wrong. Anyone reconstructing the escalation should also do the arithmetic precisely. Measured against Opus 4.7’s 7.5x launch rate (changelog), the 27x row (docs) is a 3.6x increase, and April 16 to June 1, the day multipliers went legacy, spans 46 days. For Opus 4.7 itself the documented escalation is smaller and faster: 7.5x to 15x, a doubling across 15 days.
The Counterexample: Anthropic Priced Its Exit Up Front
Anthropic’s June 9 launch of Claude Fable 5 shows what the alternative disclosure looks like, and where its limits are. The announcement priced Fable 5 and its restricted sibling Mythos 5 at $10 per million input tokens and $50 per million output tokens, which Anthropic describes as less than half the price of Claude Mythos Preview. It also pre-announced the subscription exit on Anthropic’s own plans (Pro, Max, Team, and seat-based Enterprise, not Copilot): included through June 22, removed on June 23, usage credits required after that, with the included window extended if capacity allowed (announcement). Transition terms shipped with the launch post rather than after the fact.
Reality then outran the disclosure anyway. On June 12, three days after launch, Anthropic suspended access to Fable 5 and Mythos 5 and apologized for the disruption; the models were redeployed on July 1 (update notes). GitHub mirrored the suspension across all Copilot experiences, with other Claude models unaffected (changelog). No pricing schedule covers that failure mode. Any workflow pinned to Fable 5 by name was broken for roughly three weeks regardless of plan, and the suspension notice promised restoration without giving a date.
Two details from the announcement change what a “Fable 5” request actually is, and both bear on per-request model billing generally. Queries flagged by Anthropic’s cybersecurity, biology-and-chemistry, or distillation classifiers are answered by Claude Opus 4.8 instead, a fallback Anthropic says triggers in under 5% of sessions on average (announcement). And Mythos-class traffic carries 30-day retention on both first- and third-party surfaces, excluded from training and deleted after 30 days in almost all cases (announcement). On Copilot, enabling the Fable 5 policy constitutes acknowledgment of that retention requirement, while other Claude models continue under zero data retention (changelog). Anthropic at least publishes the substitute model, the trigger rate, and the retention policy. Any model routing you depend on should answer the same three questions: which model responded, at what rate, and where the transcript went. The June 23 credit transition is covered separately.
Budgeting Against the Rate You Actually Pay
Since June 1, the right move depends on which billing system you are on.
On a monthly plan, you are token-metered. Claude Opus 4.8 lists at $5.00 per million input tokens and $25.00 per million output tokens, and 1 AI credit equals $0.01 (models and pricing). Pro+ includes $70 of monthly usage after the May 12 flex update: a $39 base matched to the subscription price, plus a $31 flex allotment GitHub says can vary over time (flex announcement). The automatic fallback model is gone, so caps and budgets are yours or your admin’s to set (billing announcement).
On a legacy annual plan, the multiplier still rules. Rebuild cost models against the current table, not the rate you integrated against: Opus 4.8 at 27x against Sonnet 4.6 at 9x (docs). Weigh that against the model freeze above, and against the plan’s end state: expiration lands in Copilot Free unless you convert, with prorated credits if you switch before then (billing announcement). For work that needs current models, the legacy plan is a dead end regardless of its multipliers.
Either way, split the routing. GitHub positioned Opus 4.7 for multi-step, tool-dependent work (launch post). Routine tool calls, context accumulation, and single-file edits rarely need the Opus line; route those turns to a cheaper model and reserve the expensive one for turns that use the capability the rate pays for. Approval gates before Opus turns and hard iteration caps bound the worst case, because an uncapped loop multiplies whatever the current rate is.
Launch Promos Are Not Budgets
The 7.5x rate was labeled promotional in the first sentence that mentioned it. The failure was interpretive. “Promotional” in vendor copy reads as introductory, and introductory reads as a stable baseline for headcount planning and CI budgets. When the standing rate landed at double the launch price, the surprise was not that GitHub raised the rate, it was that the number teams had budgeted against was never the price.
June 1 added a harder lesson: the unit itself can be retired. A budget keyed to a multiplier can be invalidated by a rate change, and also by the meter being replaced out from under it. Anthropic’s Fable 5 launch is the disciplined version of the same move, with the exit priced and dated in the launch post itself, and even that could not prevent a suspension three days in. The engineering response is identical across all three failures: treat any rate published during a launch window as marketing copy, budget against the post-promotion rate on the billing model you are actually on, and keep routing loose enough that a model can be swapped in an afternoon. Betting the launch multiplier holds is betting against the vendor’s incentive to stop subsidizing your compute.
Join the discussion
Share a useful perspective or ask a question about this article.