Can Claude Write a Construction Estimate? 12 Runs of One Bathroom Against Clamp (2026)
The same 5x8 bathroom remodel run through Claude Haiku, Opus and Fable, and the free plan, twelve times and checked line by line against Clamp: the totals, the scope, the rates and the city it guessed.
The assumptions
This model assumes
- One 5x8 bathroom, same layout, described the same way every run, no location
- Totals before markup, sales tax, contingency and permits
- Benchmark: Clamp's estimate, November 2, 2026, $13,858 direct, 62 lines
- Paid Claude answers taken November 1, 2026; free plan answers November 2, 2026
The model, step by step
Each run started a new chat with the same words and no location. Three runs each on Claude's three paid models: Haiku 5.5, the fast one; Opus 5.5; and Fable 5.1. Then three more on the free plan, logged out, which Claude answered on Sonnet 5.5.
The prompt
Ask"I'm a remodeling contractor. Write me a detailed estimate for this job with line items, quantities, unit prices, labor hours and a total direct cost before my markup." Then the job: a 5x8 bathroom, same layout; replace the 60 x 30 alcove tub in place and tile the surround to the ceiling, about 65 sq ft; a 36 in vanity with a stone top and faucet; a toilet; a new exhaust fan; two light fixtures; five accessories and a mirror; paint walls and ceiling; 28 sq ft of porcelain floor tile; plumbing and electrical stay put.
Every run gave a confident total, and every one came in under Clamp. Haiku's three were 16% apart, Opus's 13%, Fable's 6.9% and free Sonnet's 15.8%.
| Run | Total | vs Clamp |
|---|---|---|
| Run 1 (Haiku 5.5) | $13,847 | −0.1% |
| Run 2 (Haiku 5.5) | $11,933 | −13.9% |
| Run 3 (Haiku 5.5) | $13,356 | −3.6% |
| Run 1 (Opus 5.5) | $12,266 | −11.5% |
| Run 2 (Opus 5.5) | $11,760 | −15.1% |
| Run 3 (Opus 5.5) | $13,294 | −4.1% |
| Run 1 (Fable 5.1) | $11,760 | −15.1% |
| Run 2 (Fable 5.1) | $12,570 | −9.3% |
| Run 3 (Fable 5.1) | $11,922 | −14.0% |
| Free 1 (Sonnet 5.5) | $10,367 | −25.2% |
| Free 2 (Sonnet 5.5) | $11,294 | −18.5% |
| Free 3 (Sonnet 5.5) | $12,001 | −13.4% |
The prompt named no city. Eight of the nine runs priced the job for the state the account is based in anyway. Two Haiku runs priced for a specific metro in that state, one with that city's permit, and four runs charged one county's 8% sales tax.
Nobody's prompt mentioned the tub's valve or shutoffs. Each model made its own call.
| Item | Haiku | Opus | Fable |
|---|---|---|---|
| New tub valveHaiku replaced only the trim | 0 | 3 | 3 |
| Shutoff valves | 1 | 3 | 3 |
| Uncoupling membraneFable used cement board twice | 1 | 3 | 1 |
| Tile edge trim | 3 | 3 | 3 |
| Doorway transition | 2 | 3 | 2 |
Every run set its own labor rates, from $35 an hour for a helper to $115 for a plumber, and called them assumptions. At those rates the work took 90.25 to 125 labor hours. Clamp prices the same job at 79 to 98 crew hours plus subcontracted lines, at trade rates.
| Model | Hours |
|---|---|
| Haiku 5.5 | 111.5 to 125 |
| Opus 5.5 | 97.5 to 107.75 |
| Fable 5.1 | 90.25 to 95.9 |
| Sonnet 5.5 | 91.75 to 101.5 |
Claude added work to the job: GFCI and switch swaps in every Opus run, baseboard, drywall patching, and in one Fable run a $225 tile niche marked as an optional upsell. One Opus total also carried a 5% contingency, and another offered 10% on the side.
Claude is a good scope checker: ask it for every item with quantities, then ask what it missed. The prices have to come from somewhere you can trace, your own costs or rates with a date on them.
Same prompt
Every line priced
The same prompt in Clamp, and the estimate it prices: $13,858.38 at job cost, every line by trade, labor hours on the labor lines, including the valve, shutoffs and stub-outs the chatbots kept dropping. Illustration of the Clamp app.
What it means
Every line on a rate you can trace
Clamp turns a job description into an itemized draft at job cost, each line priced by Clamp with labor hours and the date its rates are from, plus what it left out.
A price you can't trace is a guess. Put the scope on your own rates.
Questions
Can Claude write a construction estimate?
It writes a detailed one, often as a spreadsheet with formulas. In twelve runs of the same 5x8 bathroom, paid totals ran −15.1% to −0.1% and free totals −25.2% to −13.4% against Clamp's estimate, on labor rates it chose and a location it guessed. Clamp prices every line from its own rates, with labor hours and the date the rates are from.
Which Claude model is best for estimates?
On this job, Fable 5.1 was the most consistent (6.9% between runs) and Opus 5.5 caught the most scope. Haiku 5.5 kept the old valve body every time and its totals were 16% apart. The free plan's Sonnet 5.5 wrote line lists up to 72 lines and landed −25.2% to −13.4%.
Why did Claude price my estimate for the wrong city?
With no location in the prompt, eight of nine runs priced the job for the state the signed-in account is based in. Name the job's town in the prompt, or check every rate against your own market.
Where does Claude get its labor rates?
It sets them itself and labels them assumptions to replace: $35 to $115 an hour across these runs. No run cited a source for a labor rate.
Sources & provenance
- Twelve Claude runs of one 5x8 bathroom estimate Clamp Research
- Clamp's estimate for the 5x8 bathroom (bath-5x8/v5, 2026-11-02) Clamp Research