A developer running Claude Max 20x, Cursor Ultra, and Copilot Pro+ pays $439 a month. Every one of those bills buys substantially the same thing: a frontier model writing code. The overlap is not a rounding error. It is most of the spend.

That stacking is the normal case now. You subscribe to Claude or ChatGPT because you want the model. You subscribe to Cursor or Copilot because you want an editor that can drive it. Then usage charges arrive on top of both. Three line items. One capability.

Here is what each vendor charges today, what those numbers mean once you look at the billing model underneath, and which part of the duplication you can actually remove.

Claude Code Pricing, Start To Finish

Claude Code is included on every Claude plan, including the free one. That matters more than it sounds. It means Claude Code pricing is just Claude subscription pricing, and there is no separate SKU to reason about.

Per Anthropic's pricing page: Free is $0 and runs at half the weekly limits for the Fable and Opus models. Pro is $20 a month billed monthly, or $17 on an annual commitment. Max comes in two sizes. $100 a month buys 5x Pro's usage; $200 buys 20x. Team seats are $25 monthly or $20 annually, with a premium seat at $125 monthly or $100 annually carrying 5x the standard seat. Enterprise is quoted per seat, with usage cost scaling by model and task.

The tier names describe throughput, not capability. Max 20x does not give you a smarter model than Pro. It gives you more room before the limit lands. That distinction decides whether upgrading is worth it.

What Codex, Cursor, And Copilot Charge

Codex follows the same pattern as Claude Code. It is not a standalone product with its own price. It rides on your ChatGPT plan, and OpenAI's documentation lists Codex access on every tier. Free is $0. Go is $8. Plus is $20. Pro is $100 for 5x limits or $200 for 20x. Business is $20 per user per month billed annually, $25 billed monthly.

Cursor sells the editor rather than the model. Per Cursor's pricing page, Hobby is free and Pro is $20 a month. Above that sit Pro+ at $60 and Ultra at $200. Teams is $40 per user per month; Enterprise is quoted. The tiers are denominated in agent request allowances, so Pro+ carries roughly 3x Pro's and Ultra roughly 20x.

Copilot is the cheapest way in. It has also moved furthest toward metered billing. GitHub's plans put Free at $0, with 2,000 completions and 50 chat requests a month. Pro is $10 per user, Pro+ is $39, Max is $100. The paid individual tiers now bundle a credit allowance rather than a request count: $15 a month of GitHub AI Credits on Pro, $70 on Pro+, $200 on Max. On the team side, GitHub's documentation lists Copilot Business at $19 per seat and Copilot Enterprise at $39.

Unstoppable Code prices the layer that drives the model, and its free tier is built to sit on top of a subscription you already hold. Per the pricing page: Free is $0. It runs on your existing Claude or Codex subscription with unlimited local workspaces and parallel agents, plus 5 AI credits a day. Pro is $19 a month, a flat-rate personal-account plan adding Remote Control from the web, access to lower-cost gateway models like DeepSeek V4 Pro, and 100 AI credits a month. Business is $29 per user per month and adds shared team workspaces, org-wide visibility into agent runs, usage reporting, and a shared credit pool. Enterprise is $49 per user per month and requires a $100,000 annual commitment.

One clarification on that last tier, because the number invites a wrong assumption. Enterprise adds no new entitlements over Business. The capability set is identical. What the commitment buys is a larger credit allotment per seat and early access, not a locked-away tier of features.

The Three Billing Models Hiding Behind Five Price Lists

Reading those numbers side by side is misleading. A $20 charge from one vendor and a $20 charge from another are not the same kind of charge. There are three models in play, and they fail differently.

Flat-rate throughput. Claude and ChatGPT sell a usage ceiling for a fixed monthly fee. You cannot overspend. You can only run out and wait. Every Claude and Codex tier works this way, which is why their ladders climb so steeply. $20 to $100 to $200 is not buying escalating intelligence. It is buying a higher wall.

Metered credits. Copilot's paid tiers and Unstoppable's credits draw down a prepaid balance as you use it. This is honest about the underlying economics, since a long agent run costs more than a completion. It is also the model where a heavy month produces a bill nobody forecast.

Per seat. Every team tier here is priced per user per month. That multiplier is the one people underestimate. Cursor Teams at $40 per user is $480 a month across twelve people, before anyone runs a single agent.

Most stacks mix all three. That is exactly why the total is hard to see. The flat-rate piece is predictable, the metered piece is not, and the per-seat piece scales with headcount rather than usage.

The Rate Limit Is The Real Price

The monthly figure is the number you negotiate over. The rate limit is the number you feel.

OpenAI publishes theirs, and they repay a close read, because they show what the tier ladder actually sells. Codex limits are measured per five-hour rolling window. Plus and Business get between 10 and 250 messages per window depending on the model. Pro 5x gets 50 to 1,250. Pro 20x gets 200 to 5,000. Anthropic's Max tiers behave the same way, resetting on a rolling window rather than daily.

Two consequences follow. Neither appears in a pricing table.

First, a mid-task wall costs more than the wait. An agent that stops halfway through a refactor also costs you the context you rebuild when you come back. That is the real product the $200 tier is selling. For someone running agents most of the working day, it can be worth it.

Second, parallel agents drain a window several times faster than one. If you have moved to running multiple agents in their own worktrees, your effective tier is lower than the one on your invoice. This is where people discover their Max 5x behaves like Pro. The fix is rarely a bigger plan. It is usually noticing which runs deserve a frontier model at all.

Why You End Up Paying Three Times

The duplication has a specific shape. Two of the three charges are avoidable. One is not.

The first bill is the model: Claude Pro, Claude Max, ChatGPT Plus, ChatGPT Pro. This one is real. Somebody has to run inference, and you will pay for it either directly or bundled.

The second bill is the runner, meaning the editor or agent environment that drives the model. Cursor, Copilot, and anything else selling you a place to work. Here is what makes it duplication rather than a second product. Cursor Pro at $20 and Copilot Pro at $10 both fold in model access you are frequently already buying from Anthropic or OpenAI separately. You pay for inference twice. Once as a subscription, once inside a tool's monthly fee.

The third bill is overage: credits, API keys, whatever your tool falls back to when the included allowance runs out. It arrives after the fact. It is the one nobody budgets for.

The second charge is the soft one. If the runner works against the subscription you already hold, it collapses to zero.

Where Unstoppable Code Fits

That collapse is what the free tier is for. Unstoppable Code connects to your existing Claude or Codex subscription and runs unlimited local workspaces and parallel agents against it, at $0. The model bill stays. The runner bill goes away.

The provider list is exactly Claude, Codex, and Ollama. The Claude and Codex runtimes ship bundled and version-pinned, so there is no separate install to keep current; Ollama is the one you supply yourself. That makes putting both frontier providers on the same task an ordinary thing to do rather than a migration you plan around.

AI Credits are an optional second path, not a required one. The model picker carries a "Use AI Credits" toggle, off by default, so chats run on your own provider subscription at your plan's flat rate. Switch it on and the run routes through Unstoppable's gateway instead, which is how you reach gateway-only models like DeepSeek V4 Pro. Worth knowing what that changes besides the bill: on that path your prompts and code transit our infrastructure on the way upstream, rather than going to your provider directly. There is also an on-quota mode that routes automatically when your own subscription hits its limit. That is the direct answer to the mid-task wall above. Credits are prepaid and overdraft is allowed, so a balance running low does not halt an agent mid-run.

Five line items, the typical stack against this one.

• Model access. Typical stack: a Claude or ChatGPT subscription. Here: the same subscription, connected directly.

• Runner. Typical stack: Cursor or Copilot, billed separately, with its own bundled inference. Here: the free tier, $0.

• Running two providers. Typical stack: two tools, two configs. Here: Claude and Codex bundled, same repo.

• Hitting a rate limit. Typical stack: wait for the window to reset. Here: on-quota routing continues the run on credits.

• Team visibility. Typical stack: per-tool dashboards. Here: an org-wide view of every agent run, usage reporting, and a shared credit pool on Business.

Paid tiers buy things the free tier does not: Remote Control from the web, the gateway models, and on Business the shared workspaces, aggregated team context, and cost analysis. Pro includes 100 AI credits each month against its $19 price, which is where the "a subscription that pays for itself" framing on the pricing page comes from.

One caution on that comparison. Business at $29 and Enterprise at $49 are per user per month. They multiply by headcount exactly like Cursor Teams and Copilot Business do. Per-seat math does not disappear at any vendor, and a five-person team should run the multiplication before reading either number as cheap.

What To Do About It

Work out your real number first. Most people have not. Add the model subscription, the editor subscription, and last month's overage. That total is what AI coding costs you, and it is usually higher than the figure people quote when asked.

Then ask which line item buys inference you are already paying for elsewhere. If your editor subscription folds in model access you hold separately, that is the duplicate. Cut it first.

If your constraint is the rate limit rather than the monthly fee, moving up a tier is a fair answer, and the $100 and $200 tiers exist for that reason. Just make the decision knowing what you are buying. More room. Not a better model.

There is a third option people skip, and it is usually the cheapest. Run fewer things against a frontier model. A rate-limit window spent on a mechanical rename is a window you do not have for the refactor that needed the reasoning.

The stack that survives is the one where you pay for inference once, and everything else is free or metered to what you actually used. Start at app.unstoppable.ai