Codex VSCode extension pricing: which plan pays for Codex in VS Code, and what each one costs

What Codex in VS Code actually costs per month

On a ChatGPT plan, nothing beyond the subscription, until you hit the window. That is the part worth understanding before you pick a tier, because the limit is a rate rather than a budget. OpenAI publishes it per model, per five hours: on Plus that is 10 to 100 GPT-5.6 Sol messages, 25 to 200 on Terra, or 250 to 2,000 on Luna. Pro at $100 multiplies those by five and Pro at $200 multiplies them by twenty.

The ranges are wide because a one line fix and a cross-file refactor both count as one message. OpenAI is direct about why it will not narrow them: "Model choice, context, reasoning, tool use, retrieval, and caching all affect usage, so prompt length alone isn't a reliable estimate." What it does publish is an average, and it is a useful anchor: GPT-5.6 usage averages 5 to 30 credits a message.

Three levers actually change your monthly consumption, and all three are in OpenAI's own guidance. Switch to Terra or Luna for routine work instead of running everything on Sol. Keep AGENTS.md small, nesting it per directory rather than shipping one enormous file into every prompt. And disable MCP servers you are not using, because each one adds context to every message you send. That last one catches people who installed six integrations in an enthusiastic afternoon.

On a Copilot Pro+ seat the arithmetic is different in kind, not degree. There you are spending AI credits, which have a published price of $0.01 each past your allowance, and Pro+ includes 3,900 base credits plus a 3,100 flex allotment. So the extension draws from the same pot as your Copilot chat and your cloud agent sessions, and when the pot runs dry you keep working and the bill grows rather than stopping. We break that mechanism down in what GitHub AI credits are, and the full plan grid sits on GitHub Copilot pricing.

Codex VSCode extension vs app: do you need both?

No, and they are billed as one thing. The ChatGPT desktop app handles local chats and is available on every plan including an API key. The extension is the same product with your editor's context wired in. They draw from the same allowance and the same five hour window, so running both costs you nothing extra beyond the messages you actually send.

The one thing neither of them does is autocomplete. Codex has no inline completion product at all, which is the cleanest functional split between it and Copilot: GitHub's completions are unlimited and never billed in credits on any paid plan. If ghost text as you type is the feature you actually want every day, Codex is not the purchase, whatever its agent is like.

So which one should you buy?

If you are one developer evaluating Codex, start on Free, then go to Plus at $20 when the window starts biting. Skip Go unless $12 a month genuinely matters, because the allowance is small enough that you will spend the difference in frustration. If you already pay for Copilot Pro+, install the Codex extension on the seat you have and decide later.

If you are buying for a team, the Business plan at $20 a user is the only route that gives you SSO and the data handling commitments, but be clear with yourself that it buys governance and larger cloud machines rather than throughput: the per-seat window is identical to a $20 Plus seat. And check the billing cadence, because $20 is the annual commitment price and month to month is $25.

If what you want is a bill that does not move, neither of these is that, and it is worth saying so plainly. This is the same trade every team now makes when it decides whether to hand a repeatable job to an agent that runs business tasks end to end or to keep paying a person to do it faster: a metered tool is cheap right up until the month somebody leans on it. Our own product takes the other side of that bet deliberately. Agentcode connects to a GitHub or GitLab repository, takes a task, edits the code, runs your tests and opens a pull request for a human to read, for a flat $29 a month with no credit balance and no five hour window. It never merges on its own.

Either way, the extension is free and the evaluation is cheap. Install openai.chatgpt, spend a fortnight on it, and read your usage dashboard before you commit to a tier. If you are still weighing the two big vendors against each other, Codex vs Copilot lays out both meters, both price ladders and what a team of ten pays on each, and how to budget for AI coding tools covers the wider problem of forecasting any of this.