The spreadsheet exists because I could not decide between two lenses and got tired of holding fourteen browser tabs open. Thirty-four 35mm primes, twenty-five standard zooms. Weight, size, price new, price used, and two letter grades (corner sharpness and build) pulled from published reviews rather than measured by me. Fifty-nine rows.
Then I ran correlations on it, which was not the plan, and found out my shortlist was wrong.
Money buys sharpness. Genuinely.
I expected this to come out mushy. Diminishing returns, brand tax, the usual story where the cheap thing is secretly fine. That is not what the numbers say for primes.
Price against corner-sharpness grade across all 34 primes gives a correlation of 0.75. That is strong. Spending more really does get you a sharper 35mm lens, reliably enough that you could nearly predict the grade from the price tag.
Zooms are looser at 0.61. Still a real relationship, but there is more slack in it: more chances to overpay, and more chances to find something underpriced. Which makes sense: a zoom has more ways to be good or bad than a prime does, and price only tracks some of them.
| Relationship | Primes | Zooms |
|---|---|---|
| Price → sharpness grade | r = 0.75 (n = 34) | r = 0.61 (n = 24) |
| Aperture → weight | r = −0.70 (n = 34) | r = −0.34 (n = 25) |
Negative aperture correlation means faster glass is heavier, because a smaller f-number is a bigger opening. For primes the tax is steep and consistent. For zooms it barely holds, since a zoom's weight is driven more by how far it zooms than by how fast it opens.
Where the cheap A-grades hide
Nine primes in the set carry an A or A− for corner sharpness. The cheapest is the Viltrox AF 35mm f/1.8 II EVO Apo at about $395 used, 355 g. Next is the Sigma 35mm f/2 DG DN Contemporary, $450, 325 g. The most expensive A-grade prime in the whole set is the Sigma 35mm f/1.2 Art at $1,050. More than twice the money for the same letter on the sharpness column, and 765 g of extra weight to carry for the privilege of f/1.2.
You are not paying for sharpness at that point. You are paying for the aperture, and the aperture is a separate purchase that happens to come bundled.
Zooms tell a similar story with a clearer bargain at the front. Twelve are graded A or A−. The cheapest is the Tamron 28–75mm f/2.8 G2 at $620 used, 540 g. The most expensive is Sony's 28–70mm f/2 GM at $2,450 and 918 g. Four times the price. One grade step, if that.
The Sony premium is real and it is not enormous
Comparing like grade against like grade, first-party against third-party, on used prices:
| Set | Grade | Sony, median | Third-party, median |
|---|---|---|---|
| Primes | A | $1,050 (n = 1) | $875 (n = 6) |
| Primes | B+ | $520 (n = 1) | $300 (n = 5) |
| Zooms | A | $2,100 (n = 2) | $1,075 (n = 2) |
| Zooms | A− | $1,015 (n = 4) | $800 (n = 4) |
| Zooms | B+ | $1,300 (n = 1) | $490 (n = 6) |
Read those sample sizes before you read the medians. Several cells are a single lens, which makes them an anecdote with a dollar sign on it, not a median. The pattern across all five rows points the same way, and the honest summary is that Sony costs somewhere between a bit more and roughly double for an equivalent sharpness grade. I would not defend a number more precise than that from this data.
Grades also cannot see autofocus reliability, firmware support, or resale liquidity, and those are exactly the things people buy first-party for. The spreadsheet has nothing to say about them.
Manual focus is not the value play
I assumed manual-focus glass would be the enthusiast's discount: give up the motor, keep the optics, save real money. The 34 primes split 23 autofocus, 11 manual, and the medians go the other way.
| Focus | Count | Median used | Median sharpness | Median build | Median weight |
|---|---|---|---|---|---|
| Autofocus | 23 | $450 | 4.20 (B+) | 4.20 (B+) | 426 g |
| Manual | 11 | $250 | 2.70 (C+) | 3.30 (B−) | 360 g |
The MF group is cheaper, but it gives up a full grade and a half of sharpness to get there. That median is dragged down by cheap manual primes (Pergear, Meike, AstrHori, all under $150) and dragged up by the Voigtländer APO-Lanthar, which is an A at $850 and one of the best lenses in the entire set. The median is describing a bimodal population, so it describes neither half well.
What survives is narrower: cheap manual focus is cheap because the glass is worse, not because you skipped the motor. Good manual focus costs about what good autofocus costs. The discount I was expecting is not there.
The bit where I was wrong
My shortlist going in was the Sony Sonnar 35mm f/2.8 ZA and the Sigma 28–70mm f/2.8 Contemporary. I picked both for size. I shoot with a small body and I resent every gram.
On size the Sonnar wins outright. 120 g, 36.5 mm long, second-shortest and second-lightest of all 34 primes, beaten only by the Samyang 35mm f/2.8 at 85 g. As a thing to carry it is essentially unbeatable.
Then I asked which primes are sharper than it and still under 300 g, and got three:
- Tamron 35mm f/2.8 Di III OSD — B+, 210 g, $210 used
- Samyang AF 35mm f/1.8 — B+, 210 g, $260 used
- Sony FE 35mm f/1.8 — B+, 280 g, $520 used
The Sonnar is a B at $430. The Tamron is a grade sharper for less than half the money and 90 g heavier. Ninety grams is a set of keys.
I have been carrying the Sonnar for two years on the argument that it is the small one, and the spreadsheet I built to justify that argument does not support it. What it supports is a much narrower claim: if the lens must disappear into a jacket pocket, 36.5 mm is the shortest that is any good, and that is worth something. If it only has to be light, I overpaid for a shape.
The Sigma zoom came out better. 470 g, fourth-lightest of the fifteen f/2.8 standard zooms, $620 used, B+ on both grades. Against the Sony 24–70 GM II it is 225 g lighter and $1,130 cheaper, and gives up one sharpness grade and the 24 mm end. That trade I would make again. The three f/2.8 zooms lighter than it either lose the tele reach (Tamron 20–40) or cost more (Sony 24–50 G), and the Thypoch 24–50 has not shipped.
Cheap manual focus is cheap because the glass is worse, not because you skipped the motor.
Method, and what to distrust
The sharpness and build columns are not measurements. They are letter grades synthesized from published reviews — Alik Griffin's lens list, phillipreeve.net, DustinAbbott, OpticalLimits, Cameralabs, DPReview, plus manufacturer specs. Reviewer consensus, converted to letters, converted by me to numbers on a scale where A = 5.0 and C− = 1.8 so that correlations could be run at all.
That conversion is the weakest link in everything above. The gaps between letter grades are not equal in reality, and I treated them as if they were. Every correlation on this page inherits that assumption. I think the direction of each finding survives it. I would not bet on the second decimal place.
Prices are ballpark US street prices as of mid-2026 — new from the usual retailers, used from MPB, KEH and eBay in good condition. Used prices move constantly and vary by more than the differences I am drawing conclusions from, in some cases.
- Character lenses grade badly on purpose. The Voigtländer Nokton Classic gets a C+ for corner sharpness and people buy it precisely for how it renders. Same for the Laowa Argus f/0.95. A clinical sharpness column cannot see the reason those lenses exist.
- Superzooms get the same unfair treatment. The 24–240 is a C+ and it is also the only lens here that covers 24 to 240.
- Sample size undercuts the Sony-versus-third-party table badly. Three of five rows have a single Sony lens in them.
- APS-C lenses are excluded, as are 16–35 ultra-wides and 70–200 teles. Standard zooms only.
- No copy-to-copy variation is accounted for anywhere. Published reviews test one sample too.
What I still do not know is whether the price-to-sharpness relationship would hold up against real MTF data instead of graded consensus. My suspicion is that it would weaken, because reviewers know what a lens costs when they review it, and that has to leak into the grade somehow. I have no way to test that with this spreadsheet, and it is the thing most likely to make the headline number wrong.