Every mini-split gets a score from 0 to 100% built from five factors, which maps onto the 5-point rating shown on the site (a 0% score would be 1.5 stars; a perfect score, 5). Each factor is scored against fixed reference points, not against the other products in the list, so adding a new model never moves another model’s rating. Mini-splits are ranked within their nominal size (9,000 to 36,000 BTU).
| Factor | Measures | Weight |
|---|---|---|
| Value | price per Btu/h of heat at 5 °F | 35% |
| Cold weather | heat kept at 5 °F, lowest heating temperature | 25% |
| Efficiency | HSPF2, SEER2, COP at 5 °F | 20% |
| Warranty | compressor and parts cover, DIY install allowed | 15% |
| Noise | quietest indoor fan setting | 5% |
Where the numbers come from
Capacities, efficiency ratings and COP are AHRI-certified ratings. They come from the NEEP cold-climate air-source heat pump list, which publishes each system’s tested heat output at 47, 17 and 5 °F. Where a system isn’t on NEEP, they come from ENERGY STAR’s certified heat pump data. Each product page links its listing and AHRI certificate number. The heating limit, noise, line set and warranty come from the maker’s spec sheet, manual or warranty terms.
Several brands sell the same system under their own names, so identical certified numbers across brands are expected. The price, line set, warranty and support still differ.
Reference points by size
| Size | Price per 1,000 Btu/h at 5 °F, scoring 0 → 100 |
|---|---|
| 9,000 and 12,000 BTU | $350 → $100 |
| 18,000 BTU | $275 → $85 |
| 24,000 and 36,000 BTU | $225 → $75 |
The factors
Value is the lowest checked price divided by the certified maximum heat output at 5 °F, per 1,000 Btu/h. Heat in cold weather is what a heat pump is hardest to buy, and two units sold as “12,000 BTU” can differ by half at 5 °F. Prices are checked by hand and dated on every page. Small units always cost more per Btu, so each size has its own scale. A unit with no certified 5 °F rating is scored on price per 1,000 Btu/h of rated cooling instead, on a scale 0.7 times as steep, and the factor is marked estimated. A unit with no current price gets the neutral middle score.
Cold weather combines two things. The first is how much of its rated 47 °F heat output the unit still makes at 5 °F (60% of the factor): 50% scores zero and 100% full marks. The second is the lowest outdoor temperature the maker says it keeps heating at (40%): 5 °F scores zero and −22 °F full marks. When only one is published, that one counts alone and the factor is marked estimated.
Efficiency is built from the certified ratings:
| Rating | Share | Scores 0 → 100 |
|---|---|---|
| HSPF2 (heating, Region IV) | 50% | 8 → 11 |
| SEER2 (cooling) | 25% | 17 → 25 |
| COP at 5 °F, at maximum output | 25% | 1.7 → 2.4 |
Heating efficiency carries the most weight because a heat pump bought for heat runs longest in winter.
Warranty combines the compressor warranty (60%: 2 years scores zero, 10 years full marks) and the parts warranty (40%: 1 year scores zero, 7 years full marks; lifetime parts cover scores full marks). Where a maker gives a self-installed unit shorter cover than a professionally installed one, the self-install terms are scored. A term the maker doesn’t state counts as the neutral middle, and the factor is marked estimated. If the maker requires a licensed installer for the warranty to apply, the factor is cut to a quarter, since a self-installed unit would then have little or none.
Noise is the quietest indoor sound level the maker publishes: 40 dB(A) scores zero and 19 dB(A) full marks. Makers measure it at different distances and fan speeds, so it carries only a small weight.
What isn’t scored
Installation type is shown but not scored. Quick-connect units install without a vacuum pump, flaring tool or gauges. Flared units need all three but often cost much less. The quick-connect list shows only the first kind. Room-size claims, Wi-Fi and line set length are listed but not scored. Independent tests are listed where they exist, but there are very few, so they aren’t scored.
Missing data
If a figure isn’t published, that factor gets the neutral middle score (50) and the product page marks it as estimated.
Ties and discontinued models
Products are ranked by score within each size. Discontinued models stay listed but rank below every current model and can be hidden with the “Hide discontinued” filter.