Stop burning your limits on the wrong model.

Tell Tierpik which AI subscriptions you have and what you're about to do. It returns the exact model and setting (effort / thinking level / mode) to pick, which meter it drains, and when to escalate.

Plan-awareSetting-level, not just modelEscalation laddersCross-app routingSavings estimate

1 · Which AI apps do you use?

Pick every one you pay for (or use free). Set the plan — it decides which models you actually have and which limits bite.

2 · What are you about to do?

Describe it in your own words, or tap a task below.

Or choose a task

3 · Tune it

How much does getting it wrong cost?

Input size

Your priority

What do you usually pick today?

Your personal cheat sheet

Every task → the model + setting to pick, for each app and plan you selected. Copy it into your notes.

Priority follows the "Your priority" setting on the Recommend tab.

Limit hit? Do this right now

Your meter map

What actually counts against what, on your plans. Most people hit limits because they don't know which meter they're draining.

The habits that save the most (ranked)

Impact estimates are directional — they come from how usage is metered (context re-reading, thinking tokens, model multipliers), not from vendor-published numbers.

Weekly budget planner

Estimate how much of your allowance a typical week eats — and how much you'd keep by following Tierpik's picks instead of always reaching for the top model. Units are relative (1 = one lightweight-model message), derived from vendor credit/API price ratios.

Which app to plan for

Your typical week (tasks per week)

Models, plans and limits — what's true right now

Compiled from vendor pricing/help pages and dated guides. Limits are compute-based almost everywhere now and vendors publish few hard numbers; treat figures as "as of the date shown". If something changed, check Settings → Usage in the app — it beats any guide.

How Tierpik decides

Minimum tier per task. Each task has a floor tier (1 light → 4 frontier) below which quality usually isn't acceptable. Stakes and input size raise the floor; "Save limits" starts one tier lower and tells you when to escalate.
Cheapest model that clears the floor — on your plan. Then the right effort / thinking level for that model.
Cost weights come from vendor credit tables and API price ratios (e.g. GPT‑6 Luna : Terra : Sol : Astra ≈ 1 : 10 : 20 : 100; Claude Haiku : Sonnet : Opus : Fable ≈ 1 : 20 : 40 : 100). Thinking/effort multiplies that.
Cross-app routing uses a capability map (e.g. live web → Perplexity/Gemini/Grok; agentic coding → Claude Code/Codex; Google Workspace → Gemini) and prefers whichever app has the least scarce allowance for that tier.
Escalation ladders are explicit: you start cheap, and you only go up when a specific symptom appears.