RE/MAX Quantum RealtyHome valueContact

How Accurate Is an Online Condo Value Estimate? What Our 1,673-Sale Test Shows (2027)

Published 2 November 2026 · By Jatin Dua, Broker of Record, RE/MAX Quantum Realty

Toronto condo skyline at blue hour with a tablet showing an abstract scatter chart (illustrative)

By Jatin Dua · Broker of Record, RE/MAX Quantum Realty · Updated 2 November 2026 · 10 min read — I tested my AI condo estimator on 1,673 GTA sales it had never seen. Here are the results by area, the honest limits and how to use them.

Short answer

In a holdout test on 1,673 GTA condo sales it had never seen, my AI condo estimator had a median error of 7.5%, and 61% of estimates landed within 10% of the actual sale price. Accuracy varies by area, from 4.1% in Islington to 13% in Toronto West, so treat it as a strong starting point and confirm with a written CMA.

This post is for Toronto and GTA condo owners who want a straight answer on how far to trust an online value estimate. I tested my own AI condo estimator on sales it had never seen, and below I share every result, area by area, including where it struggles.

How accurate is an online condo value estimate, in plain numbers?

Here are the headline results from the test of my estimator:

  • 1,673 test sales across Toronto and eight GTA cities, held back from the model while it was built;
  • 7.5% median error;
  • 61% of estimates within 10% of the actual sale price;
  • 97% of test sales matched to a specific building.

Put simply, half of the estimates landed within 7.5% of what the condo really sold for, and six in ten landed within 10%. That is a strong result for a free tool that takes about 45 seconds.

The flip side matters just as much. About 4 in 10 estimates land more than 10% from the sale price. On a $600,000 condo, 10% is $60,000. That gap is exactly why I offer every owner a written CMA. A model is a fast first answer, not the final word.

What is a holdout test, and why does it matter for a home value estimate?

A model can look brilliant on the sales it learned from and still miss on new ones. That problem is called overfitting. The fair way to measure accuracy is a holdout test: you keep a set of sales away from the model while it is being fitted, then compare its estimates with what those units actually sold for.

That is what we did. The estimator was calibrated on 22 September 2026 from REALM (TRREB MLS) sold data: 8,406 condo-apartment sales with a size range, sold between 22 March and 22 September 2026. Sales were time-adjusted to September 2026, because prices were still easing about 0.8% a month. Then 1,673 sales it had not seen became the test.

When you see an accuracy claim for any valuation tool, ask whether it was measured on unseen sales. A figure measured on the training data will always look better than real life.

What do “median error” and “within 10%” actually mean?

These are the two most common ways to describe how close an automated valuation model gets.

  • Median error is the middle miss. Line up every test estimate by how far it was from the sale price, in percent, and take the one in the middle. Half the estimates missed by less, half by more.
  • Within 10% is the share of estimates that landed within 10% of the actual sale price, above or below. The valuation industry often calls this PPE10.

Both matter. Median error tells you how close a typical estimate gets. The within-10% share tells you how often you can lean on it. A tool with a low median error but a weak within-10% share would be right on average and badly wrong too often.

How did calibration and building matching cut the error?

The accuracy came in two steps.

Step one: calibrating the area model. Every setting in the formula, from each area’s base price per square foot to the floor, view, age and condition adjustments, was fitted to the sales using robust least squares. That method stops a few odd sales from pulling the whole model off course. On the test sales, calibration cut the area model’s median error from 18% to 9%.

Step two: adding the building layer. Two units with the same size and floor can sell far apart because they are in different buildings. A building index from REALM sales between September 2025 and September 2026 covers 2,732 GTA buildings and gives each a premium or discount. Buildings with few sales have that premium shrunk toward zero, and every building is capped between −20% and +25%. This step took the median error from 8.2% to 7.5%, with 97% of test sales matched to a building.

So if you pick your building from the address list when you use the tool, you are using the sharpest version of the model. If you skip it, you fall back toward the area level.

See how close it gets for your own unit. The estimator below shows a range, a most likely figure and a confidence level in about 45 seconds.

Condo Valuation

What’s your condo
worth today?

Three quick steps. Condos don’t price like houses — your floor, your view, and whether you own parking move the number more than anything else. This weighs all of them.

01Your Building
02Your Unit
03Extras & Report

Where is the condo?

Building and area do most of the work. A Humber Bay tower and a Scarborough mid-rise are different markets entirely.

Please enter the building address or name.

Please choose the closest area.

Please choose the building age.

Tell me about your unit

Drag to your floor. In a Toronto tower each storey up is worth real money — and the view is worth more again.

Please choose your layout.

700 SQ FT
3003,000+
12
Ground
12FLOOR
160+

Mid-rise. Solid, but the premium really starts higher up.

Pick one

Extras, then your report

Parking is the single biggest add-on in a Toronto condo — in some buildings it’s worth more than a renovation.

Please choose the condition.

Please enter your name.

Please enter a valid email address.

Please enter a phone number.

No cost, no obligation.
Your details are never sold or shared.

Reading recent condo sales…

Estimated market value

—

—

$0–$0

Most likely $0 · about $0 per square foot

What moved the number

Starting from what comparable units in your area sell for, here’s what your specifics added or subtracted.

Market context

—

—
Average condo sale, your area
—
Days on market

—

Two units, same floor plan,
$90,000 apart.

That happens constantly in condos — one has the parking, the right exposure, or a board that keeps the reserve fund healthy. A model can’t see your status certificate. I can.

How this works — your estimate comes from a model built on recent Toronto & GTA condo sale data, weighting area, size, layout, floor, exposure, view, parking, locker, age and condition. It is an automated estimate for information only — not an appraisal and not a Comparative Market Analysis. Condo values also depend on the building’s reserve fund, maintenance fees, recent special assessments and status certificate, none of which a model can read. Ask me for a written CMA before you make a decision.

How accurate is the condo estimator in my area?

Accuracy is not the same everywhere. This table shows every area in the test. Read the sample note before you lean on any single row.

Area Test sales Median error Within 10% Sample note
All areas 1,673 7.5% 61% Full test
Toronto Central 588 8% 60% —
North York 222 5.3% 72% —
Toronto West 94 13% 37% —
Toronto East 25 6.1% 60% Small sample
Scarborough 110 7.2% 62% —
Humber Bay 38 6% 66% —
Islington 40 4.1% 88% —
Elsewhere in Etobicoke 49 10.9% 43% —
Mimico 8 5.4% 75% Small sample
The Queensway 4 8.9% 75% Small sample
Mississauga 152 8% 62% —
Brampton 27 9.2% 52% Small sample
Oakville 45 7.6% 53% —
Burlington 48 10.4% 48% —
Milton 16 8.8% 63% Small sample
Vaughan 94 8.1% 57% —
Richmond Hill 42 6% 69% —
Markham 71 5.7% 75% —

Source: holdout test of the AI condo estimator on 1,673 unseen REALM (TRREB MLS) condo-apartment sales, calibration of 22 September 2026. “Small sample” marks areas with fewer than about 30 test sales.

The strongest larger samples are North York (5.3% median error, 72% within 10%), Markham (5.7%, 75%) and Islington (4.1%, 88% on 40 sales). Toronto Central, the biggest group at 588 sales, sits close to the overall average.

Be careful with the small samples. The Queensway result rests on just 4 test sales and Mimico on 8. Milton has 16, Toronto East 25 and Brampton 27. A handful of sales can make an area look better or worse than it really is. For those areas, lean on the confidence level in your report and on a CMA.

Why are some areas less accurate than others?

Three areas stand out for a higher error: Toronto West (13% median error, 37% within 10%), elsewhere in Etobicoke (10.9%, 43%) and Burlington (10.4%, 48%). The test does not say why. From my own experience pricing condos, areas tend to be harder to model when they mix very different buildings: older towers next to new boutique mid-rises, lofts next to standard suites, and units whose finishes vary widely. An area base price has more work to do there, and a building with few sales cannot fully correct it.

By contrast, areas with many similar towers and plenty of recent sales give the model a clear pattern to learn. That is part of why North York and Markham test well. If you are in a harder area, my guide to condo values in Etobicoke by district shows how much prices differ from one part of a single area to the next.

What makes a single condo estimate miss by more than 10%?

When an estimate misses badly, it is usually for one of these reasons:

  • Things the model cannot read. It cannot see the reserve fund, the fee history, special assessments or the status certificate.
  • Finish and layout. A top renovation and a tired one can both be ticked “renovated”. A layout with wasted space looks the same as a smart one in square feet.
  • Unusual units. Penthouses, very large suites and two-storey lofts have few true comparables. I explain why in why online home estimates break down at the top end.
  • Thin building records. New buildings and small ones may have few sales, so the model leans on the area. My post on pricing a condo when nothing in your building has sold covers that.
  • Market drift. The model is set to September 2026 prices and is meant to be re-run quarterly.
Watch out

An accurate model on average can still be wrong for your unit. Do not set a list price, accept an offer or agree a settlement on an automated number alone. The estimate is for information only. It is not an appraisal or a CMA, and a lender or court may need a formal appraisal.

How should I use the confidence level on my estimate?

Every report shows one of three levels, and it is your best single guide to how far to trust the number.

  • High confidence means your building has a recent sales record and your area tests well. Use the range for planning with fair confidence.
  • Good confidence is the middle level. Use the range, and weigh the low end more if money is tight.
  • Indicative — confirm with a walkthrough means the building could not be matched and prices in the area vary more. Treat the figure as a rough guide only.

The range is tighter when your building has sales on record, which is another reason to pick it from the address list.

My take

I am proud of these numbers, and I publish the weak areas as well as the strong ones because you deserve to know both. A 7.5% median error on unseen sales, with building-level matching, is a strong free starting point. But I built the tool to start a conversation, not to replace one. As the tool says, a model cannot see your status certificate. I can.

How do you check an online estimate before relying on it?

  1. Check your inputs. Read “What moved the number” line by line. A wrong floor, view or parking answer can move the result more than any model error.
  2. Look at the evidence. “How this number was built” shows how many comparable sales sit behind your area and bedroom type, and the 25th percentile, median and 75th percentile prices.
  3. Compare with the market pulse. See days on market and how close to asking units sell. In a slow market, the low end of the range is often more realistic.
  4. Get a second view. A written CMA reads the documents a model cannot. For how the three kinds of value differ, see AI valuation vs a CMA vs an appraisal.

For wider context on why many owners are surprised by their number in this market, read why your Toronto condo may be worth less than you paid.

What to do next

  1. Find your area in the accuracy table and note whether it is a small sample.
  2. Run your unit through the free AI condo estimator, choosing your building from the list.
  3. Read your confidence level and the price band before the most likely figure.
  4. If you are in a higher-error or small-sample area, or your unit is unusual, ask me for a written CMA.

Need help with this? Talk to me.

I’m Jatin Dua, Broker of Record at RE/MAX Quantum Realty, and I’m the best agent to do this kind of work for you in Toronto and across the GTA. Over $100 million in GTA sales and 100+ families helped sell. Feel free to contact me any time: call or text 833-330-1925 or email [email protected].

Frequently asked questions

How accurate are online home value estimates?

Online home value estimates are usually close but not exact. My AI condo estimator for Toronto and the GTA had a 7.5% median error on 1,673 unseen 2026 sales, with 61% landing within 10% of the actual price. That means about 4 in 10 missed by more than 10%. In 2027, use an online estimate to plan and a written CMA to price.

How accurate is a condo estimate in Toronto?

In Toronto, my AI condo estimator’s accuracy varied by area in its 2026 holdout test. Toronto Central had an 8% median error on 588 sales, North York 5.3% on 222 sales and Toronto West 13% on 94 sales. Toronto East tested at 6.1%, but on only 25 sales, a small sample. For a 2027 sale, confirm with a written CMA.

What is a good median error for an automated valuation model?

Lower is better, and it must be measured on sales the model never saw. My AI condo estimator for the GTA had a 7.5% median error on 1,673 unseen 2026 sales, down from 18% before calibration. Also check the share within 10% of the sale price, often called PPE10, which was 61% for my tool.

What does within 10% mean for a home value estimate?

Within 10% means the estimate landed no more than 10% above or below the actual sale price. In my 2026 test on 1,673 unseen Toronto and GTA condo sales, 61% of estimates were within 10%. On a $600,000 condo in 2027, that means within $60,000, which is useful for planning but not for setting a price.

Why is the condo estimator less accurate in some areas?

Accuracy varies because some GTA areas mix very different buildings and have fewer similar sales to learn from. In my 2026 test, Toronto West had a 13% median error and Burlington 10.4%, while North York was 5.3% and Markham 5.7%. Areas with fewer than about 30 test sales, such as Mimico or Milton, are small samples.

Is the AI condo estimator accurate for The Queensway or Mimico?

The results look good but rest on very small samples. In my 2026 holdout test, The Queensway had an 8.9% median error on just 4 sales and Mimico 5.4% on 8 sales, both with 75% within 10%. That is too few to be sure. For a Queensway or Mimico condo in 2027, check the confidence level and ask for a written CMA.

Does picking my building make the condo estimate more accurate?

Yes. Adding the building layer cut my AI condo estimator’s median error from 8.2% to 7.5% in its 2026 test, and 97% of test sales were matched to a building. The address list covers 2,732 GTA buildings with REALM sales from September 2025 to September 2026. In 2027, a matched building also gives you a tighter range.

Can I rely on an online condo estimate to set my list price?

No. An online estimate is for information only and is not an appraisal or a CMA. Even with a 7.5% median error in testing, about 4 in 10 of my estimator’s results landed more than 10% from the sale price. In Toronto and the GTA in 2027, set your list price with a written CMA that reads your status certificate.

How often is the AI condo estimator recalibrated?

It is designed to be recalibrated every quarter. The current version was calibrated on 22 September 2026 using 8,406 REALM condo sales from March to September 2026, time-adjusted to September prices. It uses sold data, not current listings. If GTA condo prices keep easing into 2027, a September-level estimate may run slightly high.

Sources

Related reading

About the author — Jatin Dua, Toronto and GTA real estate broker

I am Jatin Dua, Broker of Record and co-founder of RE/MAX Quantum Realty Inc., Brokerage, Unit 101, 799 The Queensway, Toronto. I work with buyers and sellers across Toronto and the GTA and have helped more than 100 families sell. Four-plus years of active GTA transactions and over $100 million in sales volume.

Reach me at [email protected] or 833-330-1925, or book a call.

Please read this. General information current as at 2 November 2026. It is not legal, tax, accounting, mortgage or financial advice. I am a registered real estate broker, not a lawyer, accountant or mortgage agent. Market figures are TRREB MLS data accessed through REALM or come from the sources linked above; they describe past sales and are not a forecast. Rules, rates and programs change, so confirm details with the official source or a licensed professional before you rely on them.

More from the library

Browse every article

How this estimate is produced

The estimate blends recent selling prices for comparable units in the area with the details you entered: size, bedrooms, floor, exposure, parking, locker and condition. It is a statistical estimate, not an appraisal and not a formal opinion of value.

What it cannot see

  • Interior condition and upgrades beyond what you told it
  • Unit-specific issues: noise, an obstructed view, a pending special assessment
  • Your reserve fund and where the maintenance fee is heading
  • What is competing with you in your own building right now

What moves the number most

  • Floor and exposure, which can separate otherwise identical units
  • Parking and locker, each priced separately by buyers
  • Kitchen and bathroom condition
  • Maintenance fee per square foot against the rest of the building

Treat it as a range, not a price

A single figure to the dollar would be false precision. Use the estimate to decide whether a move is worth exploring. Before you price anything for the market, the number needs checking against what is actually selling and sitting in your building this month, and that takes a person. I will do that for you at no cost and with no obligation.

Market data is sourced from the Toronto Regional Real Estate Board. Information is deemed reliable but is not guaranteed. This tool does not provide an appraisal or a formal opinion of value, and it is not a substitute for advice on your specific property. Jatin Dua, Broker of Record, RE/MAX Quantum Realty.

Call or text 833-330-1925
Scroll to Top