🧭 Task Model Advisor
Stop guessing which model to pick.
Get the best model, context window and thinking effort for your next task —
ranked from live benchmarks, limited to the models you can actually select right now.
The problem
Your editor offers 20+ models. Opus for a user story? A flash model for a spec? Max context for a 20-line script?
You either burn money on a frontier model for trivial work, or save pennies and get a mediocre result.
Task Model Advisor does the homework for you, in one keystroke.
What you get
Pick a task, get your top 3:
Illustrative flow. Real results depend on your session models and live benchmark data.
Each recommendation shows the model, context window tier, and thinking effort, plus one sentence explaining why it fits your task.
Features
|
|
| 🎯 Task-aware |
Eight presets — write/edit code, debug, refactor, code review, specs, user stories, tests, and quick/cheap tasks — plus Other (Autre in French) for free-form tasks detected locally. Ranking uses shared engines (coding, reasoning, writing, cheap, balanced), not a separate formula per label. |
| 📊 Live benchmarks |
Artificial Analysis intelligence / coding indices and pricing, plus Arena leaderboard rankings. |
| 🔒 Only what you can use |
Never suggests a model that is not in your current session. No more "great, but I don't have access to it". |
| 💸 Cost-aware |
Frontier models don't win by default: price is part of the score, and a cheaper alternative is surfaced in the top 3 when it makes sense. |
| 🧠 Thinking & context |
Suggests a thinking effort and context tier per task, bumped for reasoning models and for "big repo" style requests. |
| ⚡ One-key workflow |
Cmd+Option+R / Ctrl+Alt+R, pick a task, pick a model, done. |
| 🛟 Always a fallback |
If the host refuses to switch models, your full config is on the clipboard as JSON. If Arena is down, ranking continues without it. |
Install
From the Marketplace (when published) — search for Task Model Advisor in the Extensions view.
From a .vsix
- Download or build
task-model-advisor-*.vsix (see CONTRIBUTING.md).
- Command Palette → Extensions: Install from VSIX…
- Select the file and reload if prompted.
Quick start
1. Add your keys. On first launch the extension adds both settings (empty) to your User settings.json — just fill them in.
| Setting |
Required |
What for |
taskModelAdvisor.artificialAnalysis.apiKey |
Yes |
Benchmarks and pricing. Get a Data API key from Artificial Analysis. |
taskModelAdvisor.cursor.apiKey |
Cursor only, optional |
Lists your Agent models. Create one in the Cursor dashboard. |
API keys live in your settings.json in plain text, like any other VS Code setting. Don't commit that file.
2. Run the command.
- Command Palette → Task Model Advisor: Recommend a model for this task
- or press
Cmd+Option+R (macOS) · Ctrl+Alt+R (Windows / Linux)
3. Pick and apply. Choose a task, choose one of the top 3, then:
| Action |
What it does |
| Apply |
Copies the config, then tries to apply it. Cursor: best-effort auto-switch. VS Code: tells you what to select. |
| Copy only |
Copies the config as JSON to the clipboard. Nothing else. |
| Refresh benchmarks |
Fetches benchmarks and session models again, then re-ranks. |
Set taskModelAdvisor.language to fr for French prompts. The default is en.
Works where you work
| Host |
Model discovery |
Validate |
| VS Code + GitHub Copilot |
Your Copilot chat models |
Copies config, shows what to select |
| Cursor |
Cursor CLI (agent), then the Cursor API if you set a key |
Copies config + best-effort switch of model, effort and Max Mode |
One package, no separate builds.
Privacy
- Network requests happen only when you run the command or hit Refresh — never in the background.
- Your custom task text (Other) is processed locally. It is never sent to Artificial Analysis, Arena or anyone else.
- No telemetry. No prompt content leaves your machine.
- Each key only goes to its own service: the Artificial Analysis key to Artificial Analysis, the Cursor key to the Cursor CLI / API.
Troubleshooting
| You see |
Why / what to do |
Artificial Analysis fetch failed (missing_api_key) |
Set taskModelAdvisor.artificialAnalysis.apiKey. |
No chat models found in this session… |
Cursor: set taskModelAdvisor.cursor.apiKey, or install the Cursor CLI and run agent login. VS Code: sign in to GitHub Copilot and open Chat once. |
Arena leaderboard unavailable… |
Expected when the mirror is down. Ranking continues with Artificial Analysis only. |
| Validate didn't change the model |
Cursor's switch is best-effort and can differ between builds. Your config is already on the clipboard — paste or apply it manually. |
Good to know
- Context window tiers (
standard / medium / high) are heuristic approximations of each host's UI labels.
- Rankings depend on third-party benchmark data. Treat them as a very good starting point, not an oracle.
- A model that cannot be linked to a benchmark is shown with the sentence
No reliable benchmark.
Go further
| |
