Type what you're about to ask your AI. Get the exact model and setting for your plan, then open it in one click.
No sign-up · Runs in your browser
MCP connector
Add Tierpik once. Then your AI checks it on its own before big tasks.
https://mcp.tierpik.com/mcp
Add the connector
Customize›Connectors›Add custom connectorPaste the URL above.
Make it automatic
Settings›Your Claude›InstructionsBefore starting any substantial task (coding, long documents, research, analysis), check Tierpik for which model and effort setting to use, and always put Tierpik's pick in the first line of your reply, even if you need more information before starting. I'm on Claude Pro.
Add the connector
Settings›Apps & Connectors›Advanced›Developer mode on›CreatePaste the URL, no authentication. On Business / Enterprise, an admin adds it.
Make it automatic
Settings›Personalization›Custom instructionsBefore starting any substantial task (coding, long documents, research, analysis), check Tierpik for which model and thinking level to use, and always put Tierpik's pick in the first line of your reply, even if you need more information before starting. I'm on ChatGPT Plus.
Add the connector
Settings›Connected Apps›Add a custom appOn gemini.google.com, paste the URL. Type @Tierpik in a prompt to use it.
Add the connector
grok.com/connectors›New Connector›CustomPaste the URL.
Add the connector
Settings›Connectors›Custom connector›RemoteName it Tierpik, paste the URL, no authentication. Turn it on in a thread.
Add a remote (HTTP) MCP server
Name it tierpik and use the URL above.
Beta: these follow each app's published steps but haven't been tested in a live chat yet.
Use the lightest model that does the job well, and step up only when you see a specific failure. Quick questions, emails, rewrites and summaries rarely need a top model; debugging, analysis and long documents need a reasoning model; only the hardest problems need a frontier model. Tierpik makes that call for your exact task and plan.
Usually three things: using the heaviest model or thinking level for everything, very long chats (every new message re-reads the whole conversation), and agent tools like Claude Code, Codex or deep research that draw from the same allowance. Lighter models for routine work and fresh chats with a short summary make the biggest difference.
As of November 2026, everyday text chat on the Instant model is listed as unlimited on every ChatGPT plan, subject to abuse guardrails. Thinking levels, GPT-6 Pro, deep research, images and the Work/Codex allowance are metered separately.
Yes. On Claude's paid plans, chat, Claude Code and other features draw from the same session and weekly limits, so a long coding session leaves less for chat.
Each task has a minimum quality tier. Tierpik picks the cheapest model on your plan that clears it, tells you the exact setting (effort, thinking level or mode) and which limit it uses, and gives you a step-up path if it falls short. Stakes and input size raise the bar; 'Save limits' starts one tier lower.
No account and no sign-up. Recommendations are computed in your browser; your app choices are saved only on your device. If you click 👍 or 👎 we record the task type, the recommended model and your vote, never the task you typed. After 👎 you can add one optional line about what was wrong; we keep it for 90 days, without your name.
Yes. Add https://mcp.tierpik.com/mcp as a custom connector in Claude, ChatGPT (developer mode), Gemini, Grok or Perplexity, then ask: "Using Tierpik, which model should I use for this?"
Every common task mapped to the model and setting to use, for one app and plan.
What to do right now, which meter you're draining, and the habits that save the most.
Ranked by how usage is metered (context re-reading, thinking, model cost), not vendor-published numbers.
How much of your allowance a typical week uses with Tierpik's picks, compared with always using the top model.
Describe one API call. Get the cheapest model that does the job, and your monthly bill.
Claude, OpenAI, Gemini and Grok APIs · prices per million tokens
MCP connector
Add Tierpik once. Then ask "which model should this call use?" while you code.
https://mcp.tierpik.com/mcp
Add the connector
claude mcp add --transport http tierpik https://mcp.tierpik.com/mcpSigned in with your Claude account and already added Tierpik in Claude? It's there already.
Add the connector
codex mcp add tierpik --url https://mcp.tierpik.com/mcpAdd to mcp.json
{
"mcpServers": {
"tierpik": { "url": "https://mcp.tierpik.com/mcp" }
}
}Add to mcp.json
{
"servers": {
"tierpik": { "type": "http", "url": "https://mcp.tierpik.com/mcp" }
}
}Add to settings.json
{
"mcpServers": {
"tierpik": { "httpUrl": "https://mcp.tierpik.com/mcp" }
}
}Add this line to your project's agent instructions
Before picking or changing the LLM model used by API calls in this project, check Tierpik (tierpik_api_recommend) for the cheapest model that does the job, and say which model it picked and why.
What each app offers on each plan, and what counts against which limit.
Add Tierpik as a connector and just ask: "Using Tierpik, which model should I use for this?". Read-only, no sign-in, nothing stored.
https://mcp.tierpik.com/mcpClaude uses a connector when a question clearly needs it, like "which model should I use?". To have it check before every big task, paste this into Settings → Your Claude → Instructions (or a project's instructions), and change the plan to yours:
Before starting any substantial task (coding, long documents, research, analysis), check Tierpik for which model and effort setting to use, and always put Tierpik's pick in the first line of your reply, even if you need more information before starting. I'm on Claude Pro.Keep Tierpik switched on in the chat's tools menu. Its advice is for your next pick: the message being answered already runs on the model you chose.
Settings → Connectors (or Connected apps) → add a custom / remote connector → paste the URL.
A tip while you type in ChatGPT, Claude, Gemini, Grok and Perplexity, one-click model switching, and live usage meters. Coming soon to the Chrome Web Store.