CLI requires extra-usage billing for fast Opus 4.6, but desktop app runs it on Max plan
Resolved 💬 2 comments Opened Apr 30, 2026 by only21mil Closed Jun 27, 2026
Summary
The desktop Claude app lets a Claude Max subscriber pick Opus 4.6 Max Fast and use it on their included Max usage. The Claude Code terminal CLI on the same OAuth account refuses with a message that fast mode requires extra-usage (pay-per-token) billing to be enabled.
This is a parity gap between clients, not a billing or auth issue.
Environment
- CLI version:
2.1.123(latest on npm at time of report) - Install method: Homebrew (
/opt/homebrew/bin/claude) - Auth: OAuth (
ANTHROPIC_API_KEYis unset, verified) - Plan: Claude Max —
oauthAccount.organizationRateLimitTier: 'default_claude_max_20x' - Org: personal org, user is admin
- Cached entitlement state in
~/.claude.json: oauthAccount.hasExtraUsageEnabled: falsecachedExtraUsageDisabledReason: 'org_level_disabled'
Repro
- Subscribe to Claude Max.
- Open the Claude desktop app — pick Opus 4.6 Max Fast from the model list. It runs on included Max usage. ✅
- Open Claude Code in the terminal, authed via the same Claude.ai OAuth account.
- Run
/model— only Opus 4.7, Sonnet, Haiku appear. No Opus 4.6 / fast option. - Run
/fast— message says fast mode requires extra-usage billing. ❌
Expected
The CLI should honor the same Max-tier fast Opus 4.6 entitlement that the desktop app honors for the same OAuth account. Either:
- Surface "Opus 4.6 Max Fast" in
/modelfor Max subscribers, or - Allow
/fastto toggle on without extra-usage billing when the account is on a Max plan.
Actual
/fast is gated on hasExtraUsageEnabled regardless of Max-tier entitlement. Forces Max users to choose between (a) using desktop, (b) enabling pay-per-token billing they don't want, or (c) not using fast Opus 4.6.
Notes
- Public docs at https://code.claude.com/docs/en/fast-mode currently say fast mode is "available via extra usage only and not included in the subscription rate limits" — but this contradicts the actual desktop-app behavior for Max subscribers, so the docs likely also need updating once the CLI catches up.
This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗