How do I find which model each call used in Claude Code?

Rius records the model on every model call in a Claude Code session, so you can read it in the session’s trace or filter the Spans tab down to calls on one model. Each session is one trace in the trace list, and the model is recorded even for folders you enable with structure only. Subagents get their own model calls, which show up with their own models too.

How it works

When you turn tracing on for a folder, the plugin from the Claude Code plugin guide turns each Claude Code session into one trace. The trace is a tree: the session, then each turn (one prompt and everything it caused), then each model call, with tool calls and subagents beneath it.

The span for each model call records the model, input and output tokens, cache reads and writes, and cost. That is what makes the calls findable:

• Model names are not content. A folder enabled with /rius:enable-here sends structure only, and that includes model names, tokens, cost and timing. You do not need /rius:enable-content-here to see which model ran.
• Subagents are separate. A subagent gets its own spans with its own model calls and tokens, nested under the call that started it. If a subagent runs on a different model than the main session, you see that in its spans.
• Every span is searchable. The trace list and the Spans tab can filter on any span attribute your sessions send, and each attribute shows how many spans carry it in the current window.

The exploring traces page covers the console views in full. For the wider picture of what the plugin does for a team, see the Coding agents guide.

Set it up

1. Install the plugin, run /rius:login, and enable the folder. The Claude Code plugin guide has the details, including installing from inside Claude Code.

   claude plugin marketplace add glassflow/rius-coding-agents && claude plugin install rius@rius-coding-agents

   Then, in Claude Code, from the repo you want traced:

   /rius:enable-here

2. Work in that repo as usual. Open the Traces page at console.rius-glassflow.com. Each session is one row and appears while it is still running. With content off, the trace is named claude-code session. If the sessions in your workspace come from several developers, click a value in the Sent by column to filter to one of them.

3. To read the model for each call in one session, open the session. The waterfall shows each model call with its token counts inline. Select a call and read the Model section: provider, requested and response model, input and output tokens, and cost.

4. To see every model call across sessions, open the Spans tab, open the All kinds menu and pick LLM. This lists one row per model call. Sort by Tokens, Cost or Duration to find the heavy ones.

5. To narrow to one model, click Add filter, pick the model attribute in the Field list (gen_ai.request.model, the attribute Rius prices calls from), choose is, and pick a value from the suggestions. Use is any of to cover several models. Attribute filters match exact values only, so choose a suggested model name rather than typing part of one.

6. To stay inside one session, use the Session link in the trace’s header. It opens Spans filtered to that session. You can also paste the session ID into the Spans search box. Add the model filter on top to see which calls in that session used which model.

7. Use Save current search… on the filter bar to keep a model filter you return to. A relative range such as the last 7 days moves with you.

8. To step through calls one at a time, use the previous and next buttons in the span panel. The shortcut ]l jumps to the next LLM span, and [l to the previous one.

You can also ask from the terminal. Run /mcp, pick rius, sign in once, and ask Claude which models your recent sessions used.

What you get

• A trace per session, listed while it runs, with the model on every model call inside it.
• A filtered view of model calls: all calls to one model, across sessions or inside one, sorted by tokens, cost or duration.
• Cost and tokens next to the model on each call. Rius prices tokens at API rates, so on a Claude Pro or Max subscription the cost measures usage and is not what you pay.
• Subagent calls visible on their own, so you can see which model a subagent used and what it cost. In long sessions, subagents can account for more than half of all tokens.
• A saved search that keeps your model filter, which you can share with everyone in the workspace.

The rows you see depend on the folder rules and sign-in on each machine. Only sessions from folders a developer enabled reach Rius.

When something else fits

If you want the model breakdown per developer or per team, the Track Claude Code usage and cost across a team guide shows how to set up one shared workspace and read usage by person. If you only want to check what the current Claude Code session is running on, that is a question for Claude Code itself, because Rius shows sessions from folders you have enabled for tracing.

Related