AI ProseLab

Same prompt. Multiple models. One verdict.

Paste the prompt you actually use — up to 200,000 words. Say what you want the writing to do. Pick up to 10 models. Every passage comes back side by side with its sentence-structure numbers, a blind assessment from Claude Opus, and what a whole novel would cost at that model's rate.

See a sample!

Sent verbatim to every model you pick. Up to 200,000 words — a whole novel, if you like. Passages come back at up to about 1,000 words.

Your success criterion, in your own words. The judge scores each passage against it.

How much each model may reason before it writes. Higher levels raise each model's line, because the quote covers the whole ceiling whether or not a model uses it.

Each model is charged at its provider's list price for the run, no markup — adding one adds only its own line. Models that fail to produce text are not charged.

Nothing is charged yet. The next page shows the price broken down by model; you can revise it or pay. Payment is by Stripe; we never see your card.

How it works

  1. You write the prompt. The real one, with your system instructions and constraints. Add a one-line goal so the judge knows what "worked" means to you.
  2. You get a quote. A flat fee based on your prompt's length, plus each model at its provider's list price, broken down line by line. Add models, remove models, adjust the lab to match your budget and needs. When you're done, click Pay and Run. Nothing is charged until then.
  3. We run it everywhere at once. Each model you picked gets the identical prompt, with the thinking level you chose, at the model's default temperature. Nothing is edited.
  4. We measure. Every passage is run through the same sentence-structure ruler: words per sentence, how short the short sentences run and how long the long ones do, fragments, sentences that open with a dependent clause, how much is dialogue, how dense the paragraphs are, and more. Real counts, no opinions.
  5. Claude Opus reads them blind. Model names hidden, order shuffled. It scores prompt compliance, prose quality, and your goal for each passage, ranks the field on all three, and names the one it would send you.
  6. You get the whole thing. Every passage in full, the numbers side by side, the verdict, and the cost of a 50,000-word novel at each model's rate. On the page, and as a PDF.

What it costs

A flat fee by prompt length, plus the models at cost. The fee covers the Claude Opus assessment and us; every model you pick is added at its provider's list price for the run, with no markup, and card processing is passed through. You see it all broken down before you pay.

Prompt of 1–25,000 words$1.00 + models cost
Prompt of 25,001–100,000 words$2.00 + models cost
Prompt of 100,001–200,000 words$3.00 + models cost
Longer than 200,000 wordsnot offered

A one-page prompt to three mid-priced models comes to about $1.60 all in. The quote is the maximum the run can cost; if a model fails to deliver, its line comes off the bill.

What this is not

It isn't a leaderboard. Rankings from a thousand generic prompts say nothing about your prompt in your voice. This runs your prompt, and the verdict is about that and nothing else.

And the judge is a reader, not an oracle. Its scores are directional; the passages are the evidence. Read them.