City bikes › Ranking

1. Real listings only

Every bike on this site was read off a German retailer's own product page during the last run (08:40 today). Nothing is hand-entered and nothing is invented. Where a retailer does not publish a fact, we say "not stated by the shop" rather than filling it in.

We ingest the whole city bike range each shop sells, €409 to €8,899 in the current dataset, and only then filter down to the budget bands shown on the homepage. The price range belongs to you, not to us.

2. Equipment value

Each bike earns points for its frame, drivetrain, wheels, brakes, a power meter if fitted, and a small adjustment for published weight (the reference for city bikes is 16.2 kg, the middle of this category: publishing a weight should neither win a bike points nor cost it any). The groupset ladder carries the most weight because it is the most reliable signal of what a city bike is worth. The point tables are the category's own, the same arithmetic calibrated on the market it is about, so a city bike is graded against the groupsets city bikes are actually built with.

Fields we could not establish score level with the cheapest known option. About a third of city bikes name a wheelset we cannot place, and putting all of them below the cheapest rim we can name would be a penalty for a shop being terse rather than for a bike being cheap. Level is the one value that neither rewards nor punishes a missing spec.

3. Value against today's market

We fit a curve across the whole live dataset, what German shops are charging right now for each equipment level, and compare each bike's asking price to that curve. The current fit uses 2405 bikes with R² = 0.63. A value score of 1.15 means the bike is priced about 15% below the middle of what German shops charge for that equipment level. A bike has to reach 1.06 before we will call it the winner of a band; below that we say there is no standout deal rather than crowning the least mediocre option.

The curve is fitted on city bikes alone. A city bike is never measured against what the market charges for a road bike, or the other way round: each category has its own inventory, its own curve and its own bar.

That score is a ranking tool, not a valuation, and we do not put it in front of you as a euro figure. Equipment explains much less of price at the top of the market than at the bottom. In today's data, bikes scoring 33 equipment points run from Cube Kathmandu Pro TR at €479.99 to Specialized Sirrus 4.0 Cactus Bloom/Mystic at €6,400, because what you are paying for there, frame prestige and aerodynamics and finishing kit, is not something we measure. So the recommendation also quotes bikes we actually found: how many bikes with equal or better equipment in your size cost more, and what the next cheapest one costs. Where we have too few comparable bikes to say anything at all, we give no value score and the bike cannot win a band.

We deliberately do not rank on discount. A 40%-off sticker tells you about the retailer's reference price, not about the bike.

The "order" control on the homepage is presentation only: it changes which order you read the bands in, never which bike won one. You can sort the cards by discount if you like. The winner of each band was still chosen on value against the market, and reordering cannot promote a bike that did not clear the bar.

4. It has to be buyable, in your size

A bike can only win a band if a frame size compatible with one you picked is confirmed in stock at that retailer, at a price we read off that retailer's own page. Where a shop prices per size, the price shown is the price of the size you can buy.

You can tick more than one size, from either scale or both: riders straddle sizes, and the same rider is an L at one brand and a 58 at the next. That widens the field rather than loosening the test: each size you pick is checked under exactly the same rules it would have faced on its own, a bike qualifies if it fits any of them, and the card tells you which one it fits. Picking three sizes never turns a letter into a centimetre we would otherwise refuse to convert.

Centimetre sizes are compared directly (±1.5 cm counts as a fit and is labelled as equivalent). A letter size is only translated into centimetres where somebody who knows the bike said so: the shop printing the manufacturer's own dual label ("XL = 58 cm"), a mapping we read out of the manufacturer's own documentation and cite (Focus Izalco Max, Focus Paralane, Giant TCR/Propel/Defy/Contend), or a ladder harvested from many such dual labels (none).

For Canyon, Trek, ROSE, Lapierre, Liv, BH, Ridley, Merida, Wilier, Colnago, Megamo, Bulls we looked and did not find enough to stand behind, so their bikes appear when you pick a letter size rather than a centimetre one. The commonest reason is that the only centimetre figure a manufacturer publishes is a tube length, which is a different thing from a size name: Giant's own chart puts its L at a conventional 58–60 and gives it a 55.5 cm seat tube. Reading the tube as the size would send you a frame two sizes out, so where we cannot tell the difference we say nothing. A wrong size is worse than a missing option.

5. The same bike at several shops

Listings are merged into one product only on brand, model, drivetrain and model year, or on a shared manufacturer barcode. Two trims, two drivetrains or two model years never merge. Where we are not certain, the listings stay separate and no "same bike cheaper here" claim is made.

6. Where AI is and is not used

Component text is parsed deterministically first. A language model is consulted only for fields that deterministic parsing could not resolve, and any value it returns is discarded unless it comes with evidence quoted verbatim from the retailer's own text. The model never chooses the winner, never decides that two bikes are the same, never converts a size, and never contributes a number to the ranking.

No sentence you read here was written by a model. The descriptive paragraph that used to sit on the bike page repeated what the spec line and the facts list above it already said; it was dropped when the site became bilingual.

The review page shows every record, every warning, every size decision and the provenance of every fact. It is in English on purpose: it is our tool, not yours.