Currently, what I see on the usage tab is a long list of…something. I’m not even sure what these line items are supposed to be. They don’t correspond to individual requests (because I see the same line item going up as an agent is working, firing off multiple requests). They don’t correspond to agent conversations, sessions, etc. either. So it’s just this big list of usage, with almost nothing contextual to help you figure out what you spent the money on. And if you’re still using your included usage, there’s nothing telling you what the cost would be once your subscription runs out. Why??
What I want to see is the cost and usage per agent (or per automation run, or per bugbot review, whatever), with the title of the agent, just as it appears in the IDE or the web interface. Break down the individual models / subagents within that agent, so I can see how many tokens were used by each, etc.
I also just don’t trust these line items. Today I fired off a review request for Fable. I saw the agent spin up subagents with different models. I was watching the usage tab while this happened, and the top line item just kept changing model name. So it would say it was GPT-5.3-codex, then it would change to an Opus model, then Fable for a bit, then back to codex. Once the task was finished, there was just a single line item for Fable, and a token count. What happened with all the subagents that were using cheaper models? Did I get billed at Fable rates for all those tokens??
And actually…did I get billed at all? The line item says “Free”, but I have no idea why, since I’ve already used my subscription allotment (which say “Included” anyway). Is there a promo for Fable such that it’s free right now? I didn’t see that in the announcement anywhere…
Hi @ryanw I’m happy to try to shed some light on the topic of usage and how it shows up in the console. There’s 3 main possibilities:
Free: Either promotional usage or in the event that a request errors out before completing successfully. You won’t be charged for requests that fail to generate a response. In your case, the Fable 5 usage coming up as free was a promotional grant. I’m not sure exactly why, but we do promotions from time to time, which could also be an individual credit on your account, etc.
Included: This is included in your subscription plan.
$x.xx: This is usage that is charged as on-demand usage/billing.
Hey @ryanw, following up specifically on the subagent part, since it’s the easiest thing to misread on the current Usage tab.
Short version: how subagents show up depends on where the agent runs, but in every case you’re billed at each model’s own rate, not the orchestrator’s.
For billing, when an agent spins up subagents on a cheaper model, every model is priced at its own rate. So if an Opus 4.8 agent fans out to three Composer 2.5 subagents, the Composer tokens are charged at Composer rates and the Opus tokens at Opus rates. Even when those costs land on a single line item, that line’s total is the true blended cost of every model that actually ran. You were not billed at Fable rates for the cheaper subagent tokens in your review. The label rolls up, the math underneath does not.
How it displays:
Local agents in the IDE: each subagent is recorded as its own usage line item with its own model name and token count. In the Opus plus Composer example, you’d see the Opus parent line plus a separate Composer 2.5 line for each subagent.
Cloud agents, background agents, and review runs: today the whole turn (parent plus subagents) collapses into one line item labeled with the orchestrator model. That’s exactly what you saw with the Fable review: the model name changing while it ran, then settling into a single Fable line at the end. The per-subagent split isn’t surfaced on those runs yet, so Composer 2.5 wouldn’t appear as its own row there.
That cloud and review rollup is a known limitation, and breaking subagent usage out into its own labeled line items is actively being worked on.
Keep in mind for your particular usage, you had a free / promotional use grant