Desktop app: No way to use Sonnet with standard (200k) context — 1M context is forced, blocks usage when credits unavailable
Resolved 💬 4 comments Opened May 25, 2026 by echase Closed Jun 29, 2026
Bug Description
On the Claude Code desktop app (macOS), Sonnet 4.6 is forced into 1M context mode with no option to select standard 200k context. When usage credits are not enabled, this results in a hard block — Sonnet is completely unusable on desktop.
Error message:
Usage limit reached — API Error: Usage credits required for 1M context · turn on usage credits at claude.ai/settings/usage, or use --model to switch to standard context
The error suggests using --model to switch to standard context, but that's a CLI flag — there is no equivalent in the desktop app UI. The /model sonnet and /model claude-sonnet-4-6 commands in the desktop app still select the 1M context variant.
Impact
- Plan: Max ($100/month)
- Severity: Complete loss of Sonnet access on desktop
- Users are paying for Max and cannot use Sonnet at all on the desktop app when 1M context credits are exhausted or not enabled
- The only workaround is switching to Opus (which uses a different rate limit bucket) or using the CLI instead
- Attempted workaround: setting
"largeContext": falsein projectsettings.local.json— no effect
Expected Behavior
- The desktop app should allow selecting standard context (200k) Sonnet, either via
/modelor a settings toggle - When 1M context credits are unavailable, the app should gracefully fall back to standard 200k context instead of blocking entirely
- The error message should not suggest CLI-only flags (
--model) as the fix when running in the desktop app
Steps to Reproduce
- Open Claude Code desktop app on Max plan
- Use
/model sonnetor/model claude-sonnet-4-6 - Send any message
- Observe: "Usage credits required for 1M context" error with no way to switch to standard context
Environment
- Claude Code Desktop (macOS, M1 Pro)
- Version: 2.1.142
- Plan: Max ($100/month)
- Model: claude-sonnet-4-6
- Context window shown: 36.4k / 1.0M (4%)
This issue has 4 comments on GitHub. Read the full discussion on GitHub ↗