groundy
developer tools

GitHub Copilot Drops Opus from Pro and Pauses Signups: The Forced Migration Facing Agentic Workflows

GitHub removed all Opus models from Copilot Pro on April 20 and flagged older versions for Pro+ removal. Opus 4.8 is the top Opus-tier model available through Copilot Pro+.

8 min···6 sources ↓

On April 20, 2026, GitHub removed all Opus models from Copilot Pro, announced that Opus 4.5 and 4.6 will also leave Pro+, and simultaneously paused new signups across all individual Copilot plans.1 Teams that built long-horizon agent workflows on Claude Opus inside Copilot Pro now face a hard choice: pay for Pro+, rewrite prompt budgets for Sonnet or Haiku, or move to direct Anthropic API billing.

What Changed on April 20

GitHub’s April 20 changelog covers three distinct moves at once1:

  • New signups paused for Copilot Student, Pro, Pro+, and Max. Existing subscribers are unaffected and can still upgrade between tiers. Two days later, on April 22, GitHub separately paused new self-serve signups for Copilot Business. [Updated June 2026]
  • All Opus models removed from Copilot Pro effective immediately.
  • Opus 4.5 and 4.6 flagged for removal from Pro+; Opus 4.7 was the only surviving Opus model on Pro+ at the time, later joined by Opus 4.8 on May 28, 2026.5

This wasn’t the first signal. On April 10, GitHub separately paused Copilot Pro free trials, citing “a significant rise in abuse of our free trial system”2, and retired Opus 4.6 Fast from Copilot Pro+ while enforcing new usage limits3. The April 20 announcement extended that contraction to the full individual plan lineup.

The Opus Squeeze: Which Models Survive on Which Tier

ModelCopilot ProCopilot Pro+
Opus 4.8RemovedAvailable
Opus 4.7RemovedAvailable
Opus 4.6RemovedRemoved
Opus 4.5RemovedRemoved
Sonnet / Haiku tiersAvailableAvailable

The net effect: if a workflow requires any Opus model, Copilot Pro is no longer a viable surface. Opus access on Pro+ now spans 4.7 and 4.8, with 4.8 being the current frontier Opus-tier model released May 28, 2026.5

What Pro+ Actually Buys You

GitHub describes Pro+ as offering “more than 5X the limits of Pro”1, but the announcement does not specify whether that multiplier applies to tokens per request, requests per month, or both. That ambiguity matters for teams running long-context agent loops: a 5x ceiling on request count is materially different from a 5x ceiling on output tokens.

Pro+ retains Opus 4.7 after the Opus 4.5 and 4.6 removals, and now also includes Opus 4.8 (released May 28, 20265), making it the only Copilot plan with access to any Opus-class model. The signup pause creates a separate problem: orgs trying to standardize on Pro+ across a growing team cannot currently provision new seats at that tier.1

As of June 1, 2026, GitHub introduced a model-multiplier billing system for annual subscribers, replacing the flat-request ceiling with per-model multipliers. Under this system, Opus 4.7 and 4.8 carry a 27x multiplier for annual request-based subscribers — meaning each Opus invocation consumes 27 premium requests. Sonnet 4.6 moved from 1x (effectively included) to 9x. [Updated June 2026] The “more than 5X the limits of Pro” marketing framing predates this multiplier table; on request-based annual billing, the effective Opus capacity under Pro+ depends on request allocation and the 27x draw-down rate, not a simple 5x ceiling increase. Teams evaluating Pro+ should model their Opus call volume against the multiplier math before committing, rather than treating the “5X” headline as a token budget comparison. For the detailed multiplier table see GitHub Copilot’s updated multiplier table and what it means for annual subscribers.

The Refund Escape Hatch (May 20 Deadline)

GitHub is offering a refund for unused subscription time to users whose plans changed.1 That window closes May 20, 2026 (30 days from the April 20 announcement).

Community discussion threads show users raising frustration about mid-billing-cycle model removals with limited notice.4 The refund window partially addresses that, but it does not restore the access that prompted the complaints.

The Signup Pause: Still Active as of June 2026

GitHub initially gave no end date for the pause and said only that it would last “until further notice.” Community discussion threads from late May suggest some users expected sign-ups to resume June 1 alongside GitHub’s broader billing restructuring. That did not happen. As of June 17, 2026, new sign-ups for Copilot Pro, Pro+, Max, Student, and Business self-serve remain paused. GitHub has stated sign-ups will reopen “in the coming weeks” but has not committed to a specific date. [Updated June 2026]

The extended pause creates a compound problem for teams trying to expand access to Pro+: they cannot provision new seats for developers who joined since April 20, which means net-new hires are either blocked from Copilot entirely or reliant on the free tier. The practical ceiling for orgs standardizing on Opus access through Copilot is the headcount they had on April 20.

GitHub’s Stated Reason vs. the Verified Anthropic Angle

GitHub’s official explanation for both the April 10 and April 20 changes points to internal infrastructure: “service reliability and a sustainable Copilot experience,” strained by “high concurrency and intense usage” on shared infrastructure.3 Neither announcement mentions Anthropic, API cost pressure, or third-party subscription policy changes.

The timing has drawn speculation. On April 4, 2026, Anthropic blocked third-party tools — starting with OpenClaw — from using Claude Pro and Max subscription quotas via OAuth, confining flat-rate subscription access to Claude.ai and Claude Code only. [Updated June 2026] That policy change is confirmed, but it targets consumer Claude subscriptions, not GitHub’s enterprise API partnership. No official announcement from Anthropic or GitHub has confirmed a causal link to the Copilot restructuring; figures for the number of affected instances circulating in community discussions remain unverified. Community threads do show users comparing Copilot’s pricing unfavorably to Anthropic’s direct API pricing4, but no participant has cited documentation establishing the causal chain.

The infrastructure-strain explanation is internally consistent without invoking external API economics: Copilot’s flat-rate model creates adverse selection for the heaviest token consumers, and Opus-class models are disproportionately expensive per token relative to Sonnet or Haiku. Whether Anthropic’s upstream pricing changes accelerated GitHub’s decision is plausible but, as of April 23, 2026, unconfirmed.

What Teams Should Do Now

The forced migration splits into three paths, each with different cost and rewrite implications.

Stay on Copilot, upgrade to Pro+. Retains the IDE integration and Opus 4.7 access, as well as the newer Opus 4.8 released May 28, 2026.5 Opus 4.8 delivers the same $5/$25 per million token pricing as 4.7 with measurable benchmark gains: SWE-Bench Pro at 69.2% versus 64.3% for 4.7, and four times less likely to allow flaws in code — see Opus 4.8 vs Opus 4.7: what changed and what did not for the full benchmark breakdown. The 5x limit increase over Pro helps with high-volume agent workloads, though the June 1 multiplier table means annual request-based subscribers should verify effective Opus capacity against the 27x per-invocation draw. The signup pause means net-new seats are currently unavailable, and orgs with growing teams hit a provisioning ceiling.1

Stay on Copilot Pro, switch models. Sonnet and Haiku remain available on Pro. For teams where Opus was used for long-context reasoning or tool-use chains, this requires auditing which tasks actually needed Opus-class performance and which were over-specified. Prompt budgets for Sonnet can be substantially leaner in token cost, which may offset the model downgrade for most inline coding tasks.

Move to direct Anthropic API billing. Removes the Copilot intermediary entirely. Direct API access gives full model selection including Opus 4.7 and Opus 4.8 (model ID: claude-opus-4-8). Claude Fable 5 (model ID: claude-fable-5), the first Mythos-class model above Opus-tier, launched June 9, 2026 at $10/$50 per million tokens6 but was suspended globally on June 12, 2026 under a US Commerce Department export-control directive; both Fable 5 and Mythos 5 remain unavailable as of June 17, 2026 while Anthropic negotiates with officials. [Updated June 2026] Per-token billing carries no flat-rate ceiling and no exposure to future IDE-tier restructuring. The tradeoff is losing native IDE integration for developers who depend on the Copilot extension’s suggestions and adding API key management overhead for every developer seat.

The structural tension the April 20 changes expose is worth naming: flat-rate IDE billing and frontier model economics do not coexist comfortably under agentic workloads. A developer using Copilot for autocomplete consumes a fraction of the tokens that an agent loop running multi-step code review or repo-wide refactoring consumes. GitHub’s tiering changes are one approach to that mismatch; direct API billing with usage caps is another. Teams that standardized on Copilot Pro for cost predictability should now price out what their actual Opus usage would cost at Anthropic API rates before assuming Pro+ is the cheaper path.

Frequently Asked Questions

Are Copilot Business or Enterprise plans affected by the April 20 changes?

The April 20 changelog is scoped to individual plans — Student, Pro, and Pro+ — so Business and Enterprise are not mentioned in that announcement. Model availability for Business and Enterprise is unchanged. However, on April 22, GitHub separately paused new self-serve signups for Copilot Business for organizations on GitHub Free and GitHub Team plans, citing the same infrastructure pressures. Existing Business customers were unaffected and can still add seats. [Updated June 2026]

Can a team that specifically needs Opus 4.6 find it on any Copilot tier?

No. Opus 4.6 (including the Fast variant) was retired from Pro+ on April 10, and the April 20 announcement removes all Opus models from Pro and flags 4.5 and 4.6 for Pro+ removal. Only Opus 4.7 survives, exclusively on Pro+. Workflows tuned for 4.6’s particular behavior must be re-validated against 4.7 regardless of which tier you pick.

What happens if a Copilot Pro subscriber does nothing before May 20?

The subscription continues unchanged in price with Sonnet and Haiku access. GitHub is not force-migrating Pro users to Pro+ or cancelling plans. The only deadline is the refund request window, after May 20, you lose the option to reclaim unused subscription time but keep the (now Opus-free) Pro plan.

Could Opus 4.7 or 4.8 also disappear from Pro+ in a future update?

The two-week cadence, Opus 4.6 Fast retired from Pro+ on April 10, then all Opus pulled from Pro and 4.5/4.6 flagged for Pro+ removal on April 20, shows GitHub narrowing model availability in iterative steps. Opus 4.7 and 4.8 on Pro+ is current-state, not contractually guaranteed. Teams betting on long-term Opus access through Copilot should maintain a parallel migration path to direct API billing. For a broader comparison of which tool provides the most durable Opus access path, see GitHub Copilot vs Cursor vs Claude Code: the 2026 AI coding showdown.

sources · 6 cited

  1. Claude Opus 4.8 — Anthropicanthropic.comvendoraccessed 2026-05-28
  2. Claude Fable 5 and Mythos 5 — Anthropicanthropic.comvendoraccessed 2026-06-10