Repository navigation
Benchmarks: 0.8.5 on Linux and macOS - #22
Conversation
The full suite with Pi-Bolt 0.8.5 from its release archives, Pi 1.1.0 as released on Bun and Node, PiG 0.4.1 and oh-my-pi 18.8.7: charts, README and BENCHMARKS.md tables, and the site's figures. Raw data in bench/results/2026-10-12-*-0.8.5.
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: 3cfc023b50
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".
| <tr><td>Writing a 200 KB file through a tool call</td><td class="n best">0.7 s</td><td class="n">38.1 s</td><td class="n">26.6 s</td></tr> | ||
| <tr><td>CPU, streaming a 60,000-character answer</td><td class="n best">6.5 s</td><td class="n">41.1 s</td><td class="n">32.6 s</td></tr> | ||
| <tr><td>Memory after a 4.2M-token session</td><td class="n best">159 MB</td><td class="n">248 MB</td><td class="n">413 MB</td></tr> | ||
| <tr><td>Writing a 200 KB file through a tool call</td><td class="n best">0.5 s</td><td class="n">38.1 s</td><td class="n">26.6 s</td></tr> |
There was a problem hiding this comment.
For Windows readers, this table presents the 0.8.5 values as directly comparable with Bun and Node, but docs/BENCHMARKS.md says the 0.8.5 figures came from a battery-powered run against 0.8.0, while the Bun/Node figures came from the older AC-powered 0.8.0 suite. Those builds were not interleaved and the power conditions differ, contradicting this page's stated methodology and making the displayed ratios unsupported. Run all three builds together or keep the 0.8.5 comparison separate.
Useful? React with 👍 / 👎.
| <text x="24.0" y="118" font-size="12.5" font-weight="650" fill="#1f2328">Pi-Bolt 0.8.0</text> | ||
| <text x="211.0" y="118" font-size="12.5" font-weight="650" text-anchor="end" fill="#1f2328">45 ms</text> | ||
| <rect x="24.0" y="125" width="26.4" height="6" rx="3" fill="#2a78d6"/> | ||
| <text x="24.0" y="62" font-size="36" font-weight="700" fill="#1f2328">38<tspan font-size="16" font-weight="600" fill="#59636e" dx="5">ms</tspan></text> |
There was a problem hiding this comment.
When the npm page loads this updated HEAD-hosted chart, sighted users see the new 0.8.5 figures, but npm/README.md line 21 still describes 0.8.0 and the old 45/128/316 ms values. Screen-reader users therefore receive a different version and measurements from the displayed image; update the npm README alt text whenever this chart changes.
Useful? React with 👍 / 👎.
The full suite again with Pi-Bolt 0.8.5 (release archives) against Pi 1.1.0 as released on Bun and Node, PiG 0.4.1 and oh-my-pi 18.8.7, on the same machines as the 0.8.0 run. Charts regenerated by bench/report.py; the README, BENCHMARKS.md and the site's home page and Benchmarks page use the same run. Windows: the laptop's 0.8.5 run against 0.8.0. oh-my-pi runs with OMP_SKIP_SETUP=1 to skip its first-start setup screen.