@calesennett/pi-codex-fast

pi extension that adds Fast and Ultrafast service tiers to supported OpenAI requests.

Packages

Package details

extension

Install @calesennett/pi-codex-fast from npm and Pi will load the resources declared by the package manifest.

npm report
$ pi install npm:@calesennett/pi-codex-fast
Package
@calesennett/pi-codex-fast
Version
0.1.10
Published
Sep 30, 2026
Downloads
734/mo · 253/wk
Author
calesennett
License
unknown
Types
extension
Size
246.4 KB
Dependencies
0 dependencies · 1 peer

Security note

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

README

pi-codex-fast

This pi extension adds Fast and Ultrafast service tiers to supported OpenAI requests.

Usage

Inside pi:

  • /codex-fast toggles Fast mode.
  • /codex-ultrafast toggles Ultrafast mode.

From CLI:

  • pi --fast
  • pi --ultrafast

You cannot enable both modes at the same time.

Persistence

The extension reads the mode from these pi settings files:

  • global: $PI_CODING_AGENT_DIR/settings.json (or ~/.pi/agent/settings.json)
  • project override: <cwd>/.pi/settings.json

Use the key pi-codex-fast.mode. The allowed values are off, fast, and ultrafast. The extension also accepts the old enabled key.

Writes go to the global settings file.

Behavior

Fast mode sets service_tier: "priority" for these models:

  • openai/gpt-5.4
  • openai/gpt-5.5
  • openai/gpt-5.6-sol
  • openai/gpt-5.6-terra
  • openai/gpt-5.6-luna
  • openai/gpt-6-astra
  • openai/gpt-6-sol
  • openai/gpt-6-luna
  • openai/gpt-6.1-sol

For each model in this list, you can also use the openai-codex/ prefix. This prefix identifies the legacy provider. Fast mode sets service_tier: "priority" with either prefix.

Ultrafast mode sets service_tier: "ultrafast" for these models:

  • openai/gpt-5.6-sol
  • openai/gpt-6-astra
  • openai-codex/gpt-6-astra

Before you use Ultrafast mode, check the Codex access requirements or the API access requirements.

The extension does not change other requests.

Example benchmark

A local live benchmark is available in this repository under evals/. Three paired trials per model produced:

Model TTFB speedup Turn speedup Wall speedup
gpt-5.6-sol 1.52x 1.58x 1.57x
gpt-5.6-terra 1.05x 1.34x 1.33x
gpt-5.6-luna 1.30x 2.31x 2.25x
gpt-6-astra 2.24x 2.59x 2.55x
gpt-6-sol 1.25x 1.07x 1.06x
gpt-6-luna 1.02x 2.02x 1.98x
gpt-6.1-sol 2.74x 1.92x 1.89x