Repository navigation
Watchable AA speed rankings, A/B compare, fullscreen zh/en demo - #2
Conversation
Ground the demo in a curated Artificial Analysis median output-speed snapshot, let ranking rows set TPS, add optional side-by-side race, fill the viewport adaptively, and consolidate persisted zh/en copy. Co-authored-by: PaintingCoder <jammyfu@users.noreply.github.com>
Constrain the 390px shell so the AA chip row and Play stay in view, and show a relative lead bar on compare panes after both streams fill. Co-authored-by: PaintingCoder <jammyfu@users.noreply.github.com>
Co-authored-by: PaintingCoder <jammyfu@users.noreply.github.com>
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ebd97d46f2
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| )} | ||
| </section> | ||
|
|
||
| <aside className="w-full min-h-0 min-w-0 max-h-[46dvh] lg:max-h-none flex flex-col overflow-hidden border-t lg:border-t-0 lg:border-l border-white/10 bg-[#0a0a0a]"> |
There was a problem hiding this comment.
At widths below lg, the sidebar is capped at 46dvh and clips overflow, but at sm widths it also reveals the full TPS, Markdown, source-toggle, and action-control stack. On short tablet or landscape viewports such as 800×600, those non-shrinking children exceed the cap and the Play/Reset controls at the bottom are clipped with no page or sidebar scrolling available. Make the sidebar internally scrollable or avoid the height cap for this breakpoint.
Useful? React with 👍 / 👎.
| <span className="w-5 shrink-0 text-[10px] font-mono text-zinc-600 tabular-nums"> | ||
| {index + 1} | ||
| </span> |
There was a problem hiding this comment.
This renders index + 1 as a rank even though MODEL_SPEED_RANKINGS explicitly contains only a curated 20-row subset of the roughly 300-row source. Consequently, positions after the first few rows claim Artificial Analysis ranks that the data does not establish, undermining the core ranking comparison. Either omit the ordinal, label it as a curated-list position, or store the actual source rank.
Useful? React with 👍 / 👎.
| label={selected ? selected.displayName : undefined} | ||
| creator={selected?.creator} | ||
| targetTps={selected?.medianTps ?? tps} | ||
| liveSpeed={streamA.currentSpeed} |
There was a problem hiding this comment.
When a desktop user changes the numeric TPS input or slider after selecting a ranking row, stream A correctly switches to the custom tps, but this header continues displaying the selected model's original AA median. Since a model is always selected and the controls are not actually locked, the simulator can therefore claim it is running, for example, Celeris-1 at 1,411.9 t/s while generating at an arbitrary user-selected rate. Clear the model selection on manual edits, lock the controls, or display the effective tps here.
Useful? React with 👍 / 👎.
2 participants
Ground Token Speed Simulator in a curated Artificial Analysis median output-speed snapshot (fetched 2026-09-13) so the ranking is the demo.
What changed
src/data/modelSpeedRankings.ts: top AA speed callouts (Celeris-1 1411.9 t/s, Mercury 2 734, Gemini 3.5 Flash-Lite 372.4, HyperNova 60B 2605 high 361) plus well-known mid/slow frontier models for contrast — 20 rows, not the full catalog.100dvhgrid: stream stage is the hero; ranking/controls sit beside it on desktop and as horizontal chips + Play under it at ~390px.src/i18n/translations.ts), persisted zh/en switcher,html langupdates.MAX_TPSraised to 2000 for AA headroom. Speeds are labeled as AA median output tokens/s — approximate for feel, not a live API benchmark.Verify
python3 tools/verify.py(lint + build) passed.document.documentElement.lang+localStorage.Note
Numbers are a static public-page snapshot. No AA API key required.
Desktop fullscreen rankings
Desktop Celeris vs Opus race
Mobile ranking chips
Mobile compare panes
aa_ranking_compare_race_language_switch_demo.mp4
To show artifacts inline, enable in settings.