[BUG] Opus 4.7 [1M] context capped at 500K on Max x20 — server-side enforcement persists across fresh install (org_level_disabled stale flag)

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

On my Max x20 plan, Opus 4.7 [1M] context window is capped at exactly
500K tokens instead of the documented 1M, while Sonnet 4.6 (1M) and
Haiku 4.5 (200K) are correctly applied on the same machine and same
login. The 500K is not a documented tier for Max plans.

The cap persists across all of the following remediation attempts:

• /extra-usage activation (no effect; this command targets billing

overflow, not context cap restoration)

• /logout → /login (no effect)
• Manual deletion of cachedExtraUsageDisabledReason line in

~/.claude.json (server immediately re-writes the line on next
launch — matches Issue #47019 pattern)

• Fresh install on a separate PC with a new login (same 500K cap

applied, conclusively confirming the issue is bound to my account
at the server side, not client-side cache)

~/.claude.json shows cachedExtraUsageDisabledReason: "org_level_disabled",
which is incorrect — my Max x20 subscription is verified active at
claude.ai/settings/billing (next renewal 2026-05-08).

Anthropic Help Center chatbot (Fin AI Agent) confirmed this is an
entitlement sync bug requiring backend reset, but cannot perform the
action and only refers ban appeals which do not apply to my case.

What Should Happen?

Opus 4.7 [1M] should provide the full 1M context window as included
in the Max x20 plan per the 1M context GA announcement:
https://claude.com/blog/1m-context-ga

Expected /context output for Opus 4.7 [1M]: xxx / 1m tokens
Actual: xxx / 500k tokens

Error Messages/Logs

/context output (Opus 4.7 [1M]):
  230.7k / 500k tokens (46%)  ← should be 230.7k / 1m

/context output (Sonnet 4.6, same machine, same session):
  102.2k / 1m tokens  ← correct (1M applied)

/context output (Haiku 4.5):
  102.2k / 200k tokens  ← correct (200K applied)

/usage output (current week):
  Current week (all models):  50% used  ← no quota exhaustion
  Current session:            55% used  ← within 5h window
  Resets Apr 30, 3pm KST

~/.claude.json flags found:
  cachedExtraUsageDisabledReason: "org_level_disabled"
  opusProMigrationComplete: true
  sonnet1m45MigrationComplete: true
  unpinOpus47LaunchEffort: true
  migrationVersion: 12
  lastReleaseNotesSeen: 2.1.119

~/.claude/backups/.claude.json.backup.* (5 files) contain:
  tengu_amber_redwood2: "500000"
(now reset to empty string in current ~/.claude.json body — suggests
 prior Statsig cohort variant push, matches `tengu_amber_*` naming
 convention; see Issue #33580 for `tengu_amber_quartz` precedent)

Steps to Reproduce

1. Sign in with a Max x20 plan account (verified active at

claude.ai/settings/billing)

1. Run claude in the terminal
2. Run /model claude-opus-4-7[1m] to select Opus 4.7 [1M]
3. Run /context — cap shown as 500k instead of 1m
4. Switch to Sonnet 4.6 with /model claude-sonnet-4-6
5. Run /context — correctly shows 1m
6. Switch to Haiku 4.5
7. Run /context — correctly shows 200k

Reproducibility: 100% on this account.
Also reproduced on a separate Windows PC with a fresh Claude Code
install and a new login session — same 500K cap, confirming
server-side account-bound enforcement.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

Same version (2.1.119) was working with 1M context until 2026-04-26 ~22:00 KST. Issue is server-side, not version-related.

Claude Code Version

2.1.119

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

Account: Max x20 ($200/month), verified active at https://claude.ai/settings/billing
(next renewal 2026-05-08).

Duration: 36+ hours since 2026-04-26 late-night KST.

Anthropic Help Center chatbot (Fin AI Agent) transcript confirmed
the diagnosis (verbatim quotes):

• "on Max plans, Opus is automatically upgraded to 1M context with

no additional configuration needed"

• "The 500K cap... isn't a documented tier for Max plans"
• "The org_level_disabled flag... suggests an entitlement sync issue

between our systems"

• "the problem likely requires backend investigation to reset the

incorrect org_level_disabled flag on your account"

The chatbot stated it cannot escalate to backend or generate a ticket
number, and referred me to the Account Ban Appeal form, which is not
applicable to my case (account is not disabled — only the entitlement
flag is incorrectly cached server-side).

Related Issues (same/similar entitlement sync pattern):

• #43872 — Max x20 blocked from Opus 1M
• #45449 — Max plan rejected from opus[1m]
• #47019 — hasExtraUsageEnabled stale, server overwrites manual edits
• #46780 — 1M billed as Extra Usage; refunds issued
• #41930 — Widespread regression since 2026-03-23, no formal communication
• #49931 — Opus 4.7 [1M] /context shows incorrect denominator
• #34178 — 1M context gated by Statsig feature flag (DISABLE_TELEMETRY

side effect)

• #33580 — tengu_amber_* Statsig feature-gate naming convention

reference

Reference: Anthropic April 23 Postmortem
(https://www.anthropic.com/engineering/april-23-postmortem)

Request:

1. Investigate the org_level_disabled flag cached against my account

and reset it. The flag is incorrect for an active Max x20 subscription.

1. Check whether my account is enrolled in any A/B test cohort

(e.g., tengu_amber_redwood2 Statsig variant) that imposes a 500K
cap, and opt me out.

1. If applicable, refund any extra-usage charges incurred during this

period (since the cap reduction may have forced workflows that
should have been covered under the Max plan).

I can provide additional logs, full /context screenshots, redacted
~/.claude.json snippets, and the chatbot transcript upon request via
private channel.