I’m on the legacy Pro plan with 500 fast requests per month (the older request-based quota plan, not the new compute credit billing).
Recently I noticed a major change in quota consumption:
Previously: One regular chat prompt usually consumed around 2 requests.
Recently: A single ordinary question frequently costs 3–6 requests.
When checking the usage dashboard, one chat message generates multiple separate billing entries, which appear to be subagent / composer background calls.
From community discussions, many legacy users are seeing the same behavior. It looks like subagent, explore and automatic tool calls are now being counted against our monthly fast request quota, without any official announcement or changelog notice.
Questions I’d like to clarify:
Has the request counting rule been updated for legacy 500-request plans recently?
Are automatic subagent / composer / explore tool calls officially counted as separate fast requests now?
Is there any toggle or setting to disable automatic background subagent calls to reduce quota consumption?
Without control over subagents, the 500-request quota drains extremely quickly and becomes unpredictable for long-time legacy subscribers.
Has anyone else on the legacy 500 plan observed the same increase in request usage?
Add a custom rule that explicitly prohibits the use of subagents. Or set a restriction so that it’s only allowed if you’ve explicitly requested it. This works without any problems.
Hey, thanks for the detailed report and screenshots. I’ll break down what’s actually happening, because your mental model is a bit off. Subagents aren’t the main reason here.
Based on your account data for the current cycle:
About 96% of the usage is claude-4.6-opus-max-thinking (a frontier model).
Only about 98 events out of ~2 700 have parent_request_id (those are real Task subagents, all on composer-2.5-fast). So subagent calls are a small share.
What actually changed: on the legacy 500-request plan, frontier models (including Opus 4.6 Thinking) require Max Mode. Enforcement was fixed on July 7. Before that, some users accidentally got 1 request per message on frontier models, and now it’s counted correctly. Details are in Colin’s staff reply here: Legacy Individual Plan - Max Mode Required (post 7).
Max Mode bills by tokens. The conversion is 4¢ = 1 request-equivalent, and one usage entry is created for every model call inside a turn (tool-loop iterations). That’s why you see 3 to 6 requests for one question. It’s not a counting bug. It’s how Max Mode works. In your screenshot, one entry with 2,91M tokens converts into 2 request-equivalents, which is expected with a large context.
Answers point by point:
The counting rule didn’t change. Only Max Mode enforcement for frontier models changed.
In Max Mode, yes, each model call, including subagents, becomes a separate token-billed entry. In non-Max legacy sessions, subagents are forced onto Composer and shouldn’t consume extra requests.
There’s no separate toggle to disable subagents. What actually helps:
Turn off Max Mode and use non-frontier models like Composer 2.5 or Auto. That gets you back to the predictable ~1 request per message.
Limit the list of subagent models in Settings > Models.
Add a user rule like Andreas suggested above, like don’t run subagents unless I explicitly ask. It’s a soft nudge, the model can ignore it, but in practice it helps.
If you want predictable 500 requests per month, the cleanest approach is to switch from Opus-Max to Composer 2.5 or Auto for daily work, and keep Opus-Max for truly heavy cases.
