How much does Codex (beta) cost per developer, and how do I measure it?

Rius prices each Codex (beta) session from its token counts at API rates and totals it per developer, so you see a dollar estimate per session and per person. There is no fixed per-developer price, because the figure follows the tokens each developer uses. On a ChatGPT plan the figure is not what you pay. It is a usage measure, priced as if the tokens were bought through the API. If a session reports no usage, its cost stays empty instead of showing $0.

How it works

Codex tracing (beta) records one trace per session. Each model call carries its model and its input, cached, output and reasoning tokens. Rius looks up the model in LiteLLM’s public price catalog and multiplies each token count by its rate. Costs are in US dollars and are list-price estimates. Rius does not apply negotiated discounts, batch pricing or long-context tiers. For the full pricing rules, see Track LLM and agent cost.

The per-developer view comes from the sign-in email:

• With plugin 0.6.2 or later, every span carries user.id, set to the email the developer used at $rius:rius-login. Sessions then appear under that person on the Users page.
• A session sent with a key stored through use-key carries no user and lands in the anonymous row.
• Codex sessions also appear as an agent named codex on the Agents page, with trace, error and cost totals.

Totals count a session only once its trace is closed. Codex does not tell plugins when a session ends, so the next Codex (beta) session on the same key closes the old trace. That happens once the old process has exited and has been idle for about an hour. Until then the session shows as in progress in the trace list. Sessions started with codex exec --ephemeral are not traced at all.

The same plugin and workspace setup works for the other coding agents. See Coding agents for the overview.

Set it up

1. Each developer installs the plugin, then opens /hooks in Codex (beta) and trusts the five rius hooks. Until they do, nothing is traced. The full install and requirements are in the Codex guide.

   codex plugin marketplace add glassflow/rius-coding-agents
   codex plugin add rius@rius-coding-agents

2. Pick one Rius workspace for the whole team so everyone’s cost adds up in one place. Each developer runs $rius:rius-login, chooses that workspace in the browser and clicks Connect. The team workflow is the same as in Track Claude Code usage and cost across a team. Codex (beta) keeps its own sign-in and folder rules.

3. In each repo you want measured, run $rius:rius-enable-here. This sends structure only: models, tokens, cost, timing and tool names, with no prompts or replies. Run $rius:rius-status to confirm tracing is on and that all five hooks are approved.

4. Work as usual. Open the Users page and sort by Cost to see who drives spend. Open a developer to see their recent and failed sessions. The trace list shows one row per session, with cost on each model call.

5. Optionally, run codex mcp login rius once. Then ask Codex (beta) in plain language: “Which of my Codex sessions this week used the most tokens?”

What you get

• A cost estimate for every Codex (beta) session, built from tokens priced at API rates, with tokens and cost on each model call.
• Cost, traces and failed traces per developer on the Users page, and the codex agent’s totals on the Agents page.
• A way to get from a costly session to the step behind it. The waterfall shows every turn, model call, tool call and subagent.
• Answers by question over the Rius MCP server, such as which sessions used the most tokens.

When a session reports no usage, the figures say so:

• If the model is not in the price catalog, or the span has no token counts, its cost is empty, not zero. In the console a total with nothing priced shows a dash.
• Averages per trace divide by every trace, priced or not, so partial coverage lowers them.
• OpenAI does not report cache-write tokens, so none are shown for Codex (beta).
• Cursor (beta) sessions carry no token counts or cost, because Cursor’s hooks do not report usage.
• Sessions from a stored key, or from plugin versions before 0.6.2, show under anonymous, not under a person.

When something else fits

• If you need the amount a ChatGPT plan actually charges you, use that plan’s own billing. The Rius figure is a list-price estimate and is not the amount you pay.
• If you need cost per developer for Cursor (beta), Rius cannot give it, because Cursor reports no usage. Claude Code and Codex (beta) report tokens and cost.

Related