How do I find which model each call used in Codex (beta)?
Rius records the model on every model call in a Codex (beta) session, so you find it by opening the session’s trace or filtering the Spans tab to LLM calls. Each Codex (beta) session is one trace, and each model call in it is its own span with the model name and its input, cached, output and reasoning tokens. Model names are sent even when a folder shares structure only, so you do not need content capture to see them.
How it works
The Rius plugin, which traces Codex in beta, turns one session into one trace. Inside it, the span tree looks like this:
- A session span at the root.
- One span per turn, which is a prompt and everything it caused.
- One span for each model call, carrying the model, input tokens, cached tokens, output tokens and reasoning tokens.
- Tool call spans and subagent spans under the model call that started them.
The model name, token counts and cost are part of the default structure-only mode. Prompts, replies and tool output are sent only from a folder you enable with $rius:rius-enable-content-here. Cost is worked out from token counts at API rates, so it is an estimate and not what you pay on a ChatGPT plan.
A session also appears as an agent named codex on the Agents page. To find model calls you work from the trace list and the Spans tab, described below. The full list of views is in Exploring traces.
Set it up
- Install the plugin. Codex (beta) support needs plugin version 0.6.0 or later.
codex plugin marketplace add glassflow/rius-coding-agents
codex plugin add rius@rius-coding-agents- In Codex, open
/hooksand trust the fiveriushooks. Until you do, nothing is traced. Then run$rius:rius-login, pick a workspace in the browser and click Connect. The full walkthrough is in Codex (beta). - In your repo, run
$rius:rius-enable-here, then$rius:rius-statusto confirm the folder is traced. Work as usual. - Open the Traces page in the console. Each session is one row, with its agent, token total, cost and latency. Set the agent filter to
codex, or click the agent cell in a row, to list only these sessions. - To list model calls across sessions, open the Spans tab and choose LLM in the All kinds menu. Sort by Tokens, Cost or Duration with the column headers.
- To narrow to one model, click Add filter, pick the requested model attribute (
gen_ai.request.model) as the Field, keep the operatorisand choose the model as the Value. Attribute filters match exact values only, so pick the value from the suggestions. Save the search with Saved if you come back to it. - To see the model on a single call, open the trace and select the model call span. Its Model section shows the provider, requested model, response model, tokens and cost.
If a session does not appear, the Hooks: line of $rius:rius-status counts the approved hooks, and the Reason line says why a folder is not traced.
What you get
- One row per Codex (beta) session in the trace list, with a Sent by column that names the owner of the key that sent it. Click it to filter to one developer, the same way you do for Claude Code usage across a team.
- A model-call list from the Spans tab, filtered by kind and by model, with tokens, cost and duration per call.
- The requested and response model on each call, so you can spot a call that ran on a different model than you expected.
- Subagent calls nested under the
spawn_agenttool call, each with its own model calls, so you can check which model a subagent used. - Sessions from every folder you enable, whether structure only or with content, searchable in the same way.
Two behaviours are useful to know. A Codex (beta) session shows as in progress until it is closed by the next session you start on the same key, once the old session’s Codex process has exited and it has been idle for about an hour. Model-call start times are approximate. For other plugin agents and the wider picture, see Coding agents.
When something else fits
- Sessions from
codex exec --ephemeralare not traced, because Codex keeps no record for the plugin to read. Use a normal session to see its model calls. - If you want the prompt or reply behind a model call, enable the folder with
$rius:rius-enable-content-here. Structure-only folders show the model but not the text.