pi-better-models

Pi extension — enhanced /models picker with Artificial Analysis coding rank & score

Packages

Package details

extension

Install pi-better-models from npm and Pi will load the resources declared by the package manifest.

npm repohome report
$ pi install npm:pi-better-models
Package
pi-better-models
Version
1.0.1
Published
Aug 19, 2026
Downloads
354/mo · 9/wk
Author
kentaylorappdev
License
MIT
Types
extension
Size
81.3 KB
Dependencies
0 dependencies · 2 peers
Pi manifest JSON
{
  "image": "https://cdn.jsdelivr.net/gh/ktappdev/pi-better-models@main/assets/model-picker.png",
  "extensions": [
    "src/extension.ts"
  ]
}

Security note

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

README

pi-better-models

A Pi extension that replaces the basic model picker with a searchable /models picker built for choosing coding models.

pi-better-models model picker showing providers, pricing, coding scores, grades, and selected-row highlighting

What you get

  • Artificial Analysis Coding Index score and letter grade (A+ to F)
  • Local rank among the models available in your current Pi session
  • Routing provider shown for every model
  • Input/output pricing
  • Optional context-window display
  • Fuzzy search
  • Tab / Shift+Tab model cycling
  • Shift+← / Shift+→ thinking-level control
  • Full-width selected-row highlighting
  • /models command plus Ctrl+L model-picker integration

Install

The package is not published on npm yet. Use the GitHub install for now:

pi install git:github.com/ktappdev/pi-better-models

Once published, the npm install will be:

pi install npm:pi-better-models

NPM package page · GitHub repository

How the score works

The picker uses the Artificial Analysis Coding Index, a 0–100 coding-focused score:

  1. It first reads benchmarks.artificial_analysis.coding from modelgrep.
  2. If modelgrep has no coding score, it uses the first-party Artificial Analysis API when an API key is configured.
  3. If neither source has a coding score, the model is shown as unscored.

There is no intelligence-score rescaling or custom heuristic. The displayed rank is calculated locally across your available, scored models.

Grades are calibrated to the current Coding Index range:

80+ A+   76–79 A    73–75 A−   70–72 B+
67–69 B  64–66 B−   61–63 C+   58–60 C
55–57 C− 50–54 D    <50 F

Optional first-party AA fallback

Modelgrep works without a key. To fill coding-score gaps with Artificial Analysis’s first-party data, set a free API key:

export ARTIFICIAL_ANALYSIS_API_KEY=aa_xxx

AA_API_KEY is also accepted as an alias. Restart Pi after changing the environment.

Choose the row details

The default compact row shows pricing and rating:

pricing · coding rank & score (AA)

Use PI_MODELS_COLUMNS to choose from context, pricing, and score:

# Default
export PI_MODELS_COLUMNS=pricing,score

# Show all fields
export PI_MODELS_COLUMNS=context,pricing,score

# Context and rating only
export PI_MODELS_COLUMNS=context,score

License

MIT © Ken Taylor