Claude Code Cost: What You Pay Solo, and What Changes on a Team
Claude Code costs $20 to $200 a month on a subscription, or per-token on an API key. It comes with Claude Pro at $20 a month, with usage limits. Max plans at $100 and $200 raise the ceiling. An API key skips the plans and bills you for what you use. And since mid-June 2026, driving Claude Code headless through the Agent SDK bills a separate metered credit pool, sold in $20, $100, and $200 blocks that don’t roll over.
That’s the solo answer. The team answer is where it gets interesting, because nobody is adding it up.
The solo math
One person, one agent, the choices are plain:
- Pro, $20/mo. Enough for regular sessions. You will hit the limit on heavy days.
- Max, $100 or $200/mo. The ceiling moves up 5x or 20x. This is where people who run Claude Code all day live.
- API key, per token. No plan, no ceiling, a bill that tracks usage exactly. Heavy agent work on a key costs real money, which is why most individuals stay on plans.
- Agent SDK credit pool. If something drives your Claude Code headless, that usage now bills separately from your plan. Metered, no rollover.
Solo, you feel your own spend. Your plan warns you. Your invoice is one line.
What changes on a team
A team running agents has a different problem: the spend goes invisible. Four people, each with their own plan or key, agents working in shared projects, some sessions running long overnight. Nobody can answer the basic question: what did this piece of work cost, and which agent spent it?
The failure mode is quiet. An agent gets stuck in a loop and burns a day of budget. A long-running session drifts. Someone’s key does work that someone else’s plan should have covered. You find out at the end of the month, or never.
Seeing it live
This is the part we built Braid to answer. Each person’s agent joins a shared channel under its own login, and the channel shows the running cost live: per-turn and per-channel totals. A turn happens, the number moves. Each owner sees their own agent’s spend, and when a room full of agents works overnight, the channel total says what it cost.
Two facts about how that works, because they’re the ones people check:
- Each agent bills its own owner. Your agent on your plan, your teammate’s on theirs. Nothing pools.
- Braid never meters your tokens. The readout is visibility, not billing. Braid’s own price is per seat and has nothing to do with your usage.
The numbers are estimates in the way all client-side cost numbers are: the agent reports what it used, and we price it against current published rates. Good enough to catch the loop that’s eating your budget. Say so plainly because a fake-precise number is worse than an honest estimate.
The free lane
If you want to try team-of-agents work without touching a paid plan at all, point the bridge at a local model through Ollama. No API key, no bill, a real agent in the channel. Slower and dumber than Claude, and free is free.
The short version
Solo Claude Code cost is a plan choice. Team Claude Code cost is a visibility problem, and the fix is watching it where the work happens. Put the agents in one channel and the number stops being a surprise.