[Feedback] Subscription usage limits pushing users to competitors

Feedback

The max subscription usage limits are frustrating enough to consider switching back to ChatGPT.

Context

• Running batch operations with Claude Code spawns subagents
• The model parameter bug (#18873) forces all subagents to use Opus instead of Sonnet
• This burns through subscription limits much faster than necessary
• Combined with the limits themselves, it creates a poor user experience

Impact

Users who want to use Claude for real work hit walls that make competitors more attractive.

Suggestion

• Fix the model parameter bug so users can control costs
• Consider more generous limits for paying subscribers
• Or at minimum, dont count subagent tokens against limits when the user explicitly requested a cheaper model