Imagine paying for your phone by the second instead of by the month. A short call would cost almost nothing. A long call would cost more than you'd guess. And on a busy day you'd start watching the clock instead of talking.
That's the choice behind how an AI teammate pays for its thinking. Socra runs each teammate on a ChatGPT subscription with Codex or a Claude subscription with Claude Code that you connect. You pay the provider a flat monthly price, and the work happens inside that plan's usage limits. The other way to run an agent is through the provider's API, where everything the model reads and writes is billed by the token. For the kind of work a teammate does all day, the subscription costs far less.
Why agent work is expensive by the token
A model doesn't remember your conversation between turns. Each time an agent takes a step, it sends the model the whole conversation so far: your instructions, the files it has opened, the commands it ran and what came back. A coding session can carry a hundred thousand tokens of context, and the agent sends it again on every step.
Providers discount context the model has already seen, but the agent still pays for it, step after step. Anthropic's current API price for Claude Opus 5.5 is $4 per million input tokens, $20 per million output tokens and $0.20 per million tokens of reused context. Each step costs little, and a teammate working for hours takes a great many of them.
A subscription prices the same work differently. Claude Pro is $20 a month and Max is $100 a month; ChatGPT Plus is $20 a month. You pay the plan price, and when you reach the plan's limit, work waits until the allowance refreshes. Extra usage costs more only if you choose to buy it.
What the same work costs both ways
The providers publish token prices and plan prices, but not a ratio between them, because it depends on how hard the agent works. People who measure their own usage give a sense of the gap.
One developer recorded every Claude Code session from June 2025 to February 2026 and priced the tokens at API rates. In the heaviest month, the work came to $5,623 at API prices on a plan that cost about $100, roughly 56 times as much. In the lightest month it was about 8 times. Over the whole period it came to more than $15,000 against roughly $800 in subscription fees, about 19 times.
These figures use the API prices of the time, and Anthropic has lowered its prices since, so the same work would cost less by the token today. In a later month, the same developer's usage was on track to cost $1,538 at API rates on a $200 plan, about 8 times. These are one person's measurements, and your ratio will depend on how much your agents work. In every month they measured, the subscription cost less than the same work by the token.
What this means for a teammate in Socra
An AI teammate in Socra has its own identity, its own computer and a role on your team, and it keeps working after you close your laptop. That long, context-heavy work is what runs up a token bill.
So Socra connects each teammate to your ChatGPT or Claude subscription instead of an API key. You don't fund an account by the token or set a spending cap on the model. The provider's price and usage limits stay separate from Socra, and teammates on the same connection share that subscription's allowance.
Not every subscription qualifies. You need a ChatGPT subscription with Codex or a Claude subscription with Claude Code, and the subscription requirements list which plans work.
Socra's own price is separate from the subscription. Socra HQ is $20 per person per month and includes $20 in shared credit. Each managed AI teammate costs $3 per month plus $1 per running hour, billed by the minute from when it wakes to when it sleeps.
See how it works at https://socra.com.