Describe the task the way you'd say it to a colleague. You'll get the exact model to open — Opus 5 or Sonnet 5, Sol or Terra, Auto or Think Deeper — the reasoning behind it, and a prompt written for your task that you can paste straight in.
Say it the way you would to a colleague — "write a proposal for a new client," "clean up this vendor list," "figure out why churn spiked." You'll get the tier, the exact model to open, the risk worth watching, and a prompt written for your task.
The whole framework is three readings. Once you can do them in your head, you won't need this tool — which is the point.
Nothing here is gated and nothing is stored but your own department and vendor, saved in this browser so the next visit picks up where you left off. Bookmark it and send it to whoever on your team keeps asking which one to use.
Most people pick one of two ways, and both cost you. They run the most powerful model available for everything — slow, expensive, and no better on routine work. Or they use whatever opened by default, get a mediocre answer on something that mattered, and quietly conclude AI isn't worth it.
Instant and inexpensive. Excellent at well-defined, repetitive work when you say exactly what you want.
Handles most real business work end to end. This is your default, and it's the correct answer more often than people expect.
Slower and pricier — and worth every bit of it on the right problem. Escalate for hard thinking or high stakes.
Fifteen departments, each with its own playbook. Then tell it whether you're on Claude, ChatGPT, Copilot, Gemini, or Grok.
Plain English, the way you'd say it out loud. No form, no jargon, no setup.
The exact model to open, why that one and not the next one up, the risk to watch — and a prompt written for your task, ready to copy.
Owners and operators paying for three AI subscriptions and unsure which one the team should open first.
Department leads who need a rule their people can follow without asking IT every time.
Finance and ops teams watching an AI bill climb without a matching climb in output.
Anyone who tried AI once, got a mediocre answer on something that mattered, and wrote it off.
A complete, board-ready B2B sales playbook tailored to your business — value props, discovery questions, buyer personas, and closing plays. Eight sections your reps can use today.
Build my free playbook →A custom Employee AI Use Policy, a ready-to-present training deck, and a signed acknowledgment memo — branded to your business and generated from your website.
Generate my policy →This tool answers a question your team asks every week, and it should have been answered once. But choosing a tier is the first ten feet of the climb. The harder work is deciding which processes are worth automating, building the systems that run them, and training your people to own what gets built — that's the work our firm does.
Match the model to the cost of being wrong. Use the fast tier when the work is well defined and a mistake is cheap to catch — tagging, extracting, reformatting. Use the balanced tier for most real business work: drafts, replies, summaries, SOPs. Escalate to the frontier tier only when a mistake is expensive or the thinking is genuinely hard — proposals, contracts, financial analysis, long documents, code, anything a client or a regulator will read.
Every major vendor — Anthropic, OpenAI, Microsoft, Google, xAI — ships a model at each of those three tiers. The tool above names the current one for yours.
They differ in speed, cost, and depth of reasoning. The fast tier is the sharp intern — instant and inexpensive, excellent at well-defined repetitive work. The balanced tier is the experienced associate, and it should be your default. The frontier tier is the senior specialist: slower and pricier, and worth every bit of it on the right problem.
No. On routine, well-defined work the flagship is slower and more expensive with no measurable gain in quality. Running it for everything is one of the two common ways teams waste money on AI — the other is using whatever opened by default on something that actually mattered.
Ask who reads the output and what a mistake costs. Anything reaching a client, a bank, a lender, a regulator, or the public is high-stakes: use the slower tier, and have a human read it before it leaves the building. Internal drafts, brainstorms, and notes aren't — treat them accordingly.
Usually, yes. Before you escalate a tier, check five things: did you attach the actual document, did you say who it's for, did you show an example of good, did you state length, tone, and format, and did you try one round of feedback. Most weak output at the balanced tier is a prompt problem — and fixing that is free.
Each vendor ships a fast, a balanced, and a frontier model, and the names change every few weeks. Pick your vendor in the tool above and it names the exact one to open, along with the catch specific to that lineup.
No. Vendors version their model families independently, so a cheap fast model can carry a higher number than the flagship sitting next to it in the same lineup. Read the tier the vendor positions the model for, not the number in its name.
For most small and mid-sized businesses, yes. Training, security review, and billing all get easier with a single default, and the capability gap between vendors at any given tier is narrower than the marketing suggests. Pick the one already embedded in the tools your team opens every day.
Never paste SSNs, bank or full card numbers, identifiable employee or medical records, customer data covered by a contract, passwords or API keys, or unreleased financials and anything under NDA. The practical test: would you hate to see it quoted back to you? Redact and generalize instead — you get the same quality of draft with none of the exposure.
Yes — free, no signup, and it works whether you're on Claude, ChatGPT, Microsoft 365 Copilot, Gemini, or Grok. Constantine Consulting publishes it alongside a Sales Playbook Builder and an AI Policy Generator.
Model names are re-checked quarterly and reflect the vendor lineups as of August 2026. The names change constantly; the framework — match the model to the cost of being wrong — is built to outlast them.
This is a judgment aid, not a guarantee. It applies a consistent framework to how you describe the task, so it's only as good as your description — and it can't see your data, your contracts, or your risk tolerance.
Model names reflect the vendor lineups as of August 2026 and change every few weeks — Anthropic, OpenAI, and Google all shipped new flagships in the eight weeks before this was written. The framework is built to outlast them; the names get re-checked quarterly. Nothing here is legal, tax, or financial advice.