Repository navigation
TUI: show real-time TPS (tokens/sec) in the bottom status bar #57275
Description
Feature request
Show tokens per second (TPS) in the OpenCode TUI bottom status bar, updated live as a model generates.
Motivation
OpenCode already reports tokens: {input, output, reasoning} per turn (visible in --format json and the API), but the TUI only shows the raw counts. A live TPS figure is the single most useful number for comparing models and spotting regressions in throughput — especially when users switch providers/models mid-session.
What's already available (no new data needed)
Each streamed text part carries a generation window:
{"type":"text","part":{"text":"...","time":{"start":1792469081557,"end":1792469083585}}}and step_finish reports tokens.output. So:
TPS = tokens.output / ((time.end - time.start) / 1000)
is computable from data OpenCode already collects. This is exactly what a small external benchmark script has to reconstruct today (see the linked discussion), which is why surfacing it in the TUI would save users the workaround.
Suggested UX
- Bottom status bar: a compact
TPS: 59.9(orTPS: 59.9 (avg 61.2)) updated onstep_finish. - Optional: dim/gray it when the model is still generating (use
time.startas the start of the current window,nowas the live estimate). - Hide it when
tokens.outputisn't reported by the provider (some providers omit usage), rather than showing 0.
Why not just the counts?
Counts alone don't tell you how fast a model is — a 120-token reply in 2s vs 10s is a very different experience, and the difference is invisible from output: 120. TPS makes the speed dimension visible at a glance, which is increasingly important now that users can switch between 20+ providers/models in one session.
Scope note
This is a TUI surfacing change only — no provider/API changes required. The data is already there; it just needs to be rendered.
Related
- External workaround: a small
opencode run --format jsonbased script that computes TPS fromtime+tokens.outputper model. Happy to share if useful.
Activity
opencode-agent commentedon Oct 20, 2026
Thanks for the detailed write-up! I think this is a duplicate of #47914, which asks for the same thing: a live tokens-per-second figure in the TUI status bar while the model generates, plus a final or average value.
Possible duplicates and related issues, ordered by likelihood:
The v2 TUI already shows tok/s for each completed turn, in the line under each assistant reply (agent · model · duration ·
N tok/s). You can turn it on or off in/settings→ Session → TPS. What's still missing is the live, status-bar version, which #47914 covers.The bot will close this issue as a duplicate in 1 day unless you reply explaining how your request is different.
LivioGama commentedon Oct 20, 2026
Closing — this is being handled locally via a CLI plugin instead of an upstream feature request.
opencode-agent commentedon Oct 21, 2026
We haven't heard back, so I'm closing this as a duplicate of #47914, which asks for the same live tokens-per-second figure in the TUI status bar. You can follow it there and add a 👍 or any details, such as your external benchmark script.
In the meantime, the v2 TUI shows tok/s for each completed turn in the line under each assistant reply. You can turn it on or off in
/settings→ Session → TPS.If your request is different from #47914, reply here and explain how, and the issue can be reopened. Thanks again for the detailed write-up!