copilot-value

Rank GitHub Copilot models by Artificial Analysis scores and workload-specific token cost

Packages

Package details

extension

Install copilot-value from npm and Pi will load the resources declared by the package manifest.

npm repohome report
$ pi install npm:copilot-value
Package
copilot-value
Version
0.2.0
Published
Sep 30, 2026
Downloads
292/mo · 12/wk
Author
el-schneider
License
MIT
Types
extension
Size
40.1 KB
Dependencies
0 dependencies · 0 peers
Pi manifest JSON
{
  "extensions": [
    "./extensions/copilot-value.js"
  ]
}

Security note

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

README

copilot-value

Ranks the models available on your GitHub Copilot subscription by Artificial Analysis benchmark score and by estimated token cost for a given workload. Useful for choosing between the dozens of Copilot models, whose prices span more than an order of magnitude and whose picker shows no quality signal.

$ copilot-value --all --top 5

#  MODEL              SCORE  CODING     USD  CREDITS  SCORE/$
1  claude-opus-5.5     57.6       -  0.6000    60.00     96.0
2  claude-sonnet-5.5   56.0       -  0.3000    30.00    186.7
3  claude-fable-5.1    53.4    81.6  1.5000   150.00     35.6
4  gpt-6-astra         52.7    76.9  1.5000   150.00     35.1
5  gpt-6.1-sol         51.8       -  0.3000    30.00    172.7

Without --all the list contains only models enabled on your subscription.

No inference calls, no quota spent, nothing changed on your account.

Install

Node.js 22.19+ and the GitHub CLI logged in (gh auth login), or GITHUB_TOKEN set.

npm install -g copilot-value

Use

copilot-value                                    # intelligence score, top 10
copilot-value --sort coding                      # coding score; only models AA scored for coding
copilot-value --sort value                       # score per dollar
copilot-value --sort price --min-score 45        # cheapest above a quality bar
copilot-value --input 200000 --cached-input 150000 --output 8000   # your workload shape
copilot-value --models gpt-6-astra,claude-opus-5 # compare a shortlist
copilot-value --json                             # for scripts and agents

--help lists everything else. Prices come from https://models.dev, scores from Artificial Analysis; both cache for 6 hours (refresh forces it). Scores are benchmarks, not your task, and cost is a token estimate, not a bill.

Intelligence is the default metric because Artificial Analysis publishes it for nearly every model. Coding scores often lag for new releases, so --sort coding lists those models under "Excluded" instead of ranking them.

Agents

Full JSON contract, caveats, and data sources: skills/copilot-value/SKILL.md.

npx skills add el-schneider/copilot-value

Users of pi get a copilot_value tool and /gh-model command with pi install copilot-value.

License

MIT