Codex Usage Dashboard: How /usage Works in 0.156.0 (2026)

For as long as Codex has existed, one question has been harder to answer than it should be: how much have I actually used? Codex usage on ChatGPT plans has always been rate-limited, but the consumption numbers lived in your plan dashboard and in the occasional throttle warning — never inside the terminal where the work happens. That changed on September 22, 2026, when Codex CLI 0.156.0 went stable with a proper codex usage dashboard: type the /usage command in any session and you get an analytics view of account usage, token totals, and plugin and skill activity, in place.

It is not the only thing that shipped. The same release added an optional fullscreen interface (the codex tui everyone on the issue tracker kept asking about), turned voice conversations on by default, enabled worktrees out of the box, and added a /daemon command for managing the local background server. Roughly 300 pull requests from 75 contributors went into 0.156.0, and the alpha line kept moving the same day — 0.157.0-alpha.5 through alpha.10 all landed within hours of the stable tag.

This post is the field guide: how to open each new panel, what the numbers mean, and what to do when the dashboard confirms what you suspected — that your subscription is the bottleneck. Spoiler from the last section: the same agentic workflow runs on DeepSeek's API for pennies, and you can try it in the browser before touching a config file.

Quick answer

• Check your usage: type /usage in any Codex CLI session — account usage, token totals, plugin and skill activity.
• Fullscreen UI: type /tui — the new interface loads on your next launch, with transcript search, mouse selection, and right-click copying.
• Voice: on by default now. F8 toggles it, /voice opens the settings picker.
• Background server: /daemon updates the local background server; --no-daemon bypasses it.
• The numbers look ugly every day? The DeepSeek API starts at $0.15 per million input tokens off-peak — the wiring guide is at the bottom.

What shipped in Codex CLI 0.156.0

If you are on an older build, none of these commands exist yet — the official changelog's update path is npm install -g @openai/codex@0.156.0. The sandbox fixes alone justify the upgrade: closed isolation gaps around inbound Windows connections, privileged Linux/macOS sockets, and writes through read-only macOS file handles.

The /usage command: reading the codex usage dashboard

Open any interactive Codex session and type /usage. That is the whole invocation — no flags, no config. The release notes describe the result as an analytics dashboard where you can "explore account usage, token totals, and plugin and skill activity."

What the panel tracks

Three buckets, straight from the release notes, with the practical reading of each:

• Account usage — where your consumption stands at the account level. If you share a plan with teammates, this is the aggregate view rather than your personal slice.
• Token totals — the raw token burn. This is the number to watch if you are trying to understand why a heavy refactor week looks different from a light maintenance week, and the natural place to check Codex CLI usage before you commit to a bigger plan.
• Plugin and skill activity — which extensions and skills are consuming what. This one is quietly the most useful: an agent that delegates to skills can burn far more than your own prompts suggest, and until now there was no per-component visibility at all.

One honest caveat: OpenAI's release notes name the buckets but do not document every widget, so treat the panel itself as the detail. What is unambiguous is the direction — check codex token usage in the terminal, in real time, without leaving your session.

/daemon and the local background server

The same release formalizes something that had been semi-visible for a while: Codex can run a local background server, and 0.156.0 gives you a command to manage it. Run /daemon to update that background server through a menu; add --no-daemon when launching if you would rather bypass it entirely and keep everything in the foreground process.

When does the distinction matter? --no-daemon gives CI and scripted runs no shared background state to clean up; /daemon keeps the background component current for interactive users without reinstalling the CLI. Neither is required knowledge — but when something behaves oddly, knowing which mode you ran is half the diagnosis.

Codex TUI: the fullscreen interface

The second headline feature answers the oldest complaint about coding agents in terminals: the interface itself. Type /tui and Codex switches to an optional fullscreen UI on your next launch — the release is explicit that it is opt-in and applies going forward, so your current session stays as-is until you relaunch.

What you get inside, per the release notes:

• Transcript search — find the command, diff, or decision from two hours ago without scrolling back through a mile of terminal buffer. For long agentic sessions this alone changes the experience.
• Mouse selection — select text the way you do everywhere else, instead of negotiating with your terminal emulator's keyboard shortcuts.
• Right-click copying — copy straight from the selection, no pipe-to-clipboard detours.

Alongside the interface, 0.156.0 adds six new themes and renders supported Mermaid diagrams and display equations directly in responses — so an architecture sketch or a bit of math arrives drawn, not as a wall of graph TD source.

How different is this from the old interface? The default inline experience still exists and still works; /tui is an alternative mode, not a replacement you were migrated onto. The practical test is a long session: if you have ever lost a plan or a file listing to scrollback and given up looking for it, the fullscreen transcript search is worth the relaunch. tmux residents get a bonus: this release also fixed clipboard forwarding over tmux/SSH — together the two changes retire the classic copy-paste workarounds.

Voice is on by default now

The third change most people will notice within a minute of updating: voice conversations are enabled by default. The mechanics, per the release notes: F8 toggles voice on and off mid-session, /voice opens a settings picker, and the CLI now bundles audio runtimes for Linux and Windows — the runtime ships with Codex rather than being a separate setup scavenger hunt.

The same release ships a voice reliability fix: speech is no longer dropped during playback pauses or audio bursts — the classic symptom of voice cutting out mid-sentence during long dictations.

If voice was something you explicitly turned off in an earlier build, check your settings after updating — the default flipped. And if you want the full picture (microphone choice, dictation quality on real refactors, where voice actually saves time versus typing), our Codex voice mode tutorial covers the whole workflow.

What the dashboard tells you about Codex limits

The usage panel answers "how much have I used," not "how much do I get." Codex usage on ChatGPT plans has always been rate-limited, and 0.156.0 does not change anyone's allowance — what it changes is visibility, which is honestly the more useful half. Once you can see token totals per session, the levers get obvious:

1. Model choice. The same day 0.156.0 went stable, OpenAI put GPT-6 Sol and GPT-6 Luna into Codex (codex --model gpt-6-sol or codex --model gpt-6-luna), with the changelog noting token pricing below the GPT-5.6 predecessors. Lighter jobs on Luna stretch the same allowance further.
2. Timing. Batch the heavy agentic runs instead of dribbling them across the day, and watch the account usage graph tell you whether it worked.
3. Scope. The plugin and skill activity view exists precisely so you can spot the component that quietly doubles your burn.

Meanwhile the lineup itself is churning — GPT-5.5 retires from Codex on October 14, 2026, with OpenAI pointing users at Sol. If reading a usage dashboard mostly taught you that your subscription feels tight, that is not a configuration problem; it is a market signal. We compared the replacements when Spark was retired in our Codex Spark alternatives roundup, and the cheap end of that list is where the next section goes.

The near-free route: DeepSeek

Here is the section for anyone whose /usage panel reads like a countdown timer. DeepSeek's API is pay-per-token at prices that make a flat subscription hard to justify for light and medium use ($ per 1M tokens, per DeepSeek's pricing docs):

deepseek-flash carries a 1M-token context window with 384K max output, and off-peak pricing roughly halves the bill. The kind of session that visibly dents a Codex allowance costs cents on this meter.

To be precise about the wiring: Codex CLI itself is built around OpenAI's models, so the honest way to run the same agentic workflow on DeepSeek is a terminal agent that accepts an Anthropic-compatible endpoint. That is a two-variable swap:

export ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic
export ANTHROPIC_API_KEY=sk-your-deepseek-key

Our DeepSeek harness setup guide walks that end to end, and the Claude Code router tutorial shows the router-based setup when you want to switch models without editing environment variables. Prefer a free ride first? Cline's desktop app includes free access to a DeepSeek flash model — the DeepSeek in Cline tutorial covers it.

And if the whole terminal dance is the part you would rather skip: our browser tools run DeepSeek models with zero setup — start at DeepSeek, or go straight to the DeepSeek App Builder to ship a working web app from one prompt. Check the usage dashboard after your Codex session, check the browser tab after ours, and compare.

FAQ

How do I check my Codex usage? Type /usage in any Codex CLI session (0.156.0 or later). The analytics dashboard shows account usage, token totals, and plugin and skill activity. On older versions the command does not exist — update with npm install -g @openai/codex@0.156.0 first.

What does the Codex /usage dashboard show? Three buckets, per the official release notes: account-level usage, token totals, and plugin and skill activity. OpenAI has not documented every widget in the panel, so the exact charts vary — but those three data sets are what the command is documented to surface.

What is the Codex TUI and how do I enable it? The TUI is the optional fullscreen interface added in 0.156.0, with transcript search, mouse selection, and right-click copying. Type /tui in a session; the fullscreen UI loads on your next launch. Your existing inline interface remains the default — nothing was forced on anyone.

How do I turn off Codex voice mode? Press F8 to toggle it off in-session, or open /voice to adjust the settings. Voice ships enabled by default starting with 0.156.0, so if you had it off before the update, re-check after upgrading.

What is /daemon in Codex CLI? /daemon manages the local background server Codex can run — the release notes describe using it to update that server. Launching with --no-daemon bypasses the background server entirely, which is handy for CI runs or scripted usage where shared background state is unwelcome.

What is the cheapest alternative when Codex limits run out? The DeepSeek API: deepseek-flash starts at $0.15 per million input tokens off-peak with a 1M-token context — a heavy agent session costs cents, not a plan upgrade. Wire it into Claude Code-compatible agents with the two environment variables above, try it free inside Cline's desktop app, or use our browser-based tools with no setup at all.

Related reading

• Codex voice mode tutorial — the full voice workflow now that it ships enabled by default
• Codex Spark alternatives roundup — seven replacements compared when OpenAI retired its speed-first model
• DeepSeek harness setup guide — wire any terminal agent to DeepSeek's API end to end
• Claude Code router tutorial — switch models through a local router without env-var edits
• DeepSeek in Cline tutorial — the free DeepSeek flash model inside Cline's desktop app
• Gemini CLI free alternative — the same near-free story on Google's terminal agent