Pick a small, honest scope.
New wired IEMs sold in India at or under ₹10,000; Type-C and 3.5mm ranked separately; sold-out kept unless discontinued.
The two plugs are ranked separately because they are different products at different prices. 210 of the415 were in stock on 2026-11-11.
Score sound with one published model.
Sound is the Harman in-ear predicted preference model and nothing else. It is a peer-reviewed formula that predicts how much listeners will like an IEM's tuning from its measured frequency response.
100.08 - 8.5*SD - 6.80*|slope| - 3.48*mean_abs_error (error vs Harman IE target, 20 Hz-10 kHz)
Three things cost points: an uneven response, a tilt away from the target, and plain distance from it. Try it.
Showing CCA PLA13.
- Harman score
- 78.3
- After the margin of error
- 74.8
- Sound points, out of 60
- 41.8
Sound points scale the adjusted score from 40 (zero points) to 90 (full points). A proxy measurement loses 5 more.
One model, one caveat. In the Harman research, about 72% prefer Harman-like tuning; about 28% prefer ~2 dB less bass and 1-2 dB more treble. That group gets advice, not a separate ranking: favour IEMs with a slightly positive tilt.
Use measurements other people already made.
The frequency responses come from public squig.link databases, compiled by the open-source Squig-Rank project. All of them use 711-type couplers. I recompute the score from three numbers per measurement: unevenness, slope and mean error.
- Each measurermay measure one IEM several times. Take the median.
- Across measurerstake the median again. One loud outlier cannot swing the score.
- One score per IEMwith a count of who measured it, and credit.
How many people measured each IEM
| Measurers | IEMs |
|---|---|
| 1 | 172 |
| 2 | 81 |
| 3 | 46 |
| 4 | 18 |
| 5 | 15 |
| 6 | 9 |
| 7 | 3 |
| 8 | 11 |
| 9 | 1 |
| 10 | 1 |
| 11 | 2 |
| 12 | 1 |
Most IEMs rest on one or two measurers. That is why step 09 exists. A further 55 have none.
Leave out the sources that said no.
Some of the best-known sources are missing on purpose. Each one either keeps its data private or asks not to be collected.
- Amazonterms forbid scraping
- HiFiNagerobots.txt blocks AI crawlers
- Crinacle / graph.hangout.audioraw data not public
- Crinacle ranking list, squig.link lists, IEMRanking.comrobots.txt blocks Claude
- Redditbulk collection needs approval; research outside its programme not allowed
I respected every site's wishes.
Build a review pipeline. Then delete it.
- Store reviewsI built a fake-review detector: verified-buyer weighting, duplicate text, bursts, imported reviews.
- Enthusiast opinionReddit now requires approval for API access and forbids research outside its programme.
- Expert lists and forumsThey covered only about 10% of the IEMs. Opinions could not be scored fairly.
- DecisionMeasurements plus store facts only. No star ratings anywhere on this site.
Take store claims as claims.
If a store says an IEM has a "woody sound", that is recorded as a claim and never verified. The same goes for specs and box contents. Everywhere you see them on this site, they carry this label:
as claimed by the store
Specs, sound tags and box contents are as claimed by the store, not verified.
Let two AI agents argue the formula out.
I did not want to hand-pick weights and call it science. So one agent proposed a scoring formula and a second one attacked it. Both could use only the data. It converged in 2 rounds, about 22 minutes.
Round 1
- Your "no mic" flag is a store-formatting artifact. Stores disagree with each other about the same IEM.
- Your discontinued rule is circular. Sold-out rows vouch for each other.
- Your "value" metric is secretly a price ranking.
Round 2
The exchange above is condensed. The three fixes are real and each one changed the final formula.
Set default weights. Then check what they really do.
The defaults are sound 60, hardware 30, in the box 10. But a weight is not the same as influence. Sound scores vary a lot between IEMs; box contents barely do.
| Part | Weight | Real influence on the ranking |
|---|---|---|
| Sound | 60 | 74 to 78% |
| Hardware | 30 | 16 to 25% |
| In the box | 10 | 2 to 7% |
The thin line is the weight. The thick bar is the share of the ranking it actually drives.
Sensible changes to the weights barely change the top 20. That is itself a finding. It only breaks when sound stops being the main ingredient.
| Sound / hardware / box | 3.5 mm | Type-C |
|---|---|---|
| 80 / 15 / 5 | 17 of 20 stay | 17 of 20 stay |
| 70 / 20 / 10 | 18 of 20 stay | 18 of 20 stay |
| 50 / 35 / 15 | 16 of 20 stay | 17 of 20 stay |
| 34 / 33 / 33 | 7 of 20 stay | 14 of 20 stay |
Build the doubt into the score.
Independent measurers disagree on the same IEM by a standard deviation of 4.94 points. So every score is reduced by a margin of error, √(24.4 ÷ measurers). One measurer costs 4.9 points. Four cost 2.5.
Do more measurers agree more?
Show this chart as a table
| Measurers | IEMs | Median spread |
|---|---|---|
| 2 | 81 | 3.5 |
| 3 | 46 | 8.4 |
| 4 | 18 | 5.6 |
| 5 | 15 | 9.5 |
| 6 | 9 | 9.4 |
| 7 | 3 | 13.0 |
| 8 | 11 | 11.6 |
| 9 | 1 | 19.8 |
| 10 | 1 | 16.9 |
| 11 | 2 | 7.1 |
| 12 | 1 | 16.4 |
- Proxy A collab edition where only the base model is measured loses 5 more points.
- Buyer beware An IEM with no measurement is scored at the bottom-quarter level of measured IEMs, 66.9. Unknown should not outrank known.
- Noise Rigs, eartips and insertion depth change measurements; treat +/-3 points as noise.
Refuse to fit a curve that is not there.
The usual way to find value is to fit a price curve and see who beats it. Here the R² of Harman score against log price is 0.001 for 3.5 mm and 0.008 for Type-C. There is no curve to beat.
So value is simple and says what it is:
percentile(score) - percentile(ln price), per plug, measured IEMs (in stock or sold out) in the top half by score only
Do not punish a store for writing less.
Some stores list every accessory. Others write two lines. If a missing spec counted as "no", the product would sink because of the store's copywriting. So a missing item gets the average rate for that item, not zero.
- Yesfull points
- Nozero
- Not statedthe average rate
Remove an IEM only on a real signal.
Sold out is not the same as discontinued. An IEM is removed only if there is a clearance tag or a newer version in stock. 19 were removed. Everything else that is sold out stays, with a flag.
See the 19 removed IEMs
| IEM | Plug | Why | Harman | Last price |
|---|---|---|---|---|
| 7Hz Aero | 3.5 mm | store clearance tag | 84.8 | ₹890 |
| KZ x Angelears Libra | 3.5 mm | store clearance tag | 83.8 | ₹990 |
| Tripowin x HBB Kailua | 3.5 mm | store clearance tag | 83.1 | ₹3,990 |
| Kiwi Ears Canta | 3.5 mm | store clearance tag | 79.8 | ₹5,699 |
| Aful MagicOne | 3.5 mm | store clearance tag | 78.5 | ₹8,990 |
| Juzear 41T | 3.5 mm | store clearance tag | 78.2 | ₹9,490 |
| Kinera Celest Relentless | 3.5 mm | store clearance tag | 76.3 | ₹7,990 |
| NiceHCK F1 Pro | 3.5 mm | store clearance tag | 75 | ₹5,990 |
| JoyOdio Shine | 3.5 mm | store clearance tag | 72.9 | ₹3,490 |
| Kinera Celest Plutus Beast | 3.5 mm | store clearance tag | 72.2 | ₹6,490 |
| Twistura Sigma | 3.5 mm | store clearance tag | 72.1 | ₹4,490 |
| ND Nice | 3.5 mm | store clearance tag | 70.8 | ₹1,290 |
| Rose Technics QuietSea | 3.5 mm | store clearance tag | 68.8 | ₹3,490 |
| Final E500 | 3.5 mm | store clearance tag | 65.5 | ₹1,499 |
| Final E500 | Type-C | store clearance tag | 65.5 | ₹2,499 |
| Audiocular LEX | Type-C | store clearance tag | 64.1 | ₹1,149 |
| ND DD3 | 3.5 mm | store clearance tag | 59.5 | ₹790 |
| KZ ZAT | 3.5 mm | store clearance tag | 55.2 | ₹5,990 |
| Tanchjim Ola | 3.5 mm | newer version in stock: Tanchjim Ola II | 48 | ₹3,699 |
'Sold out, status unclear' items may in fact be discontinued; there is no stock history.
Make every number reproducible.
- Collect politelyRate-limited requests to public store feeds from Headphone Zone, The Audio Store and Concept Kart.
- Read the pages455 product pages read by parallel AI agents for specs and box contents.
- Match and scoreEach product is matched to its measurements, then scored by script.
- PublishThree JSON files. This site reads them and nothing else.
Raw frequency-response curves are not redistributed; only derived numbers, with credit to each measurer.