copilot-value

Get the most out of your GitHub Copilot subscription.

• Which model gives me the most for my AI credits? copilot-value
• Which model is strongest for coding? copilot-value best

Copilot offers about 30 models. The same task can cost 100× more on one than on another, and the model picker shows no quality score. copilot-value ranks the models your plan enables by benchmark score and token cost.

$ copilot-value
Best value · 30 models on your plan · LMArena WebDev Elo
Each numbered row scores higher and costs more than the one above. Indented: alternatives within 50 points in the same price tier.

#  MODEL               SCORE    COST
1  gpt-6-luna           1583  $0.015
2  gemini-3.7-flash     1593  $0.113
     gemini-3.8-flash   1581  $0.113
3  grok-4.7             1636  $0.260
4  claude-sonnet-5.5    1699  $0.300
     gpt-6-sol          1692  $0.300
     kimi-k3            1659  $0.450
5  claude-opus-5.5      1820  $0.600

Cost per task: 100k input, 10k output.
18 more models omitted: each is beaten on price and score by a listed model (copilot-value best lists all).
Not ranked: claude-opus-4.8-fast, gpt-5-mini, gpt-6.1-sol, mai-code-1.1-flash (--verbose for reasons).
Scores: LMArena WebDev leaderboard (https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset), CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/), filtered and ranked by model-frontier · Prices: models.dev (https://models.dev)

Read-only: 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.

Token order: COPILOT_GITHUB_TOKEN, GH_TOKEN, GITHUB_TOKEN, then the gh login. With several gh accounts on one host, --user LOGIN uses that account's token and ignores the token variables. Copilot accepts gh OAuth tokens and fine-grained PATs with the Copilot Requests permission, but no classic PATs (ghp_...). A classic PAT in GH_TOKEN or GITHUB_TOKEN is skipped with a notice. --all ranks the published catalog and needs no token.

npm install -g copilot-value

Use

copilot-value                        # best value for money
copilot-value best                   # best models for coding
copilot-value --input 200000 --cached-input 150000 --output 8000   # your workload shape
copilot-value --margin 0             # strict frontier, no alternatives
copilot-value --all-versions         # include older models of a listed family
copilot-value best --models claude-opus-5.5,gpt-6-sol   # compare a shortlist
copilot-value --host corp.ghe.com --user alice_corp   # an enterprise or second gh account
copilot-value --verbose              # plus benchmark variants, exclusion reasons, timestamps
copilot-value --json                 # for scripts and agents

--help lists everything else. Scores are benchmarks, not your task, and cost is a token estimate, not a bill.

Scores

Works without any key: scores come from the public LMArena WebDev leaderboard (human-preference Elo on coding tasks, CC BY 4.0).

Set ARTIFICIAL_ANALYSIS_API_KEY (free) to use the Artificial Analysis Intelligence Index instead. Arena then acts as a second opinion: a model it rates close to the frontier also shows up as an alternative, tagged arena. --source aa|arena picks the ranking source explicitly. Prices come from https://models.dev. Everything is cached for 6 hours; copilot-value refresh forces a fetch.

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. Both check eligibility with pi's own Copilot login, including GitHub Enterprise logins, not with gh or token variables.

License

MIT