Duels
A weighted score tells you how a tool performs in isolation. A duel tells you which one people actually pick when they can only have one. The two answers disagree often enough to be worth publishing separately, and every split below is the raw vote count.
8 duels settled
Claude Code
51.9%
41 votes
Cursor
48.1%
38 votes
79 votes cast. Claude Code takes the duel by 3 votes.
The coding agent final 51.9 percent to 48.1 percent over 79 votes, the tightest result in the category. Claude Code takes the long task. Voters consistently report better retention across a refactor that spans an afternoon and better beh
Lovable
53.1%
43 votes
Bolt.new
46.9%
38 votes
81 votes cast. Lovable takes the duel by 5 votes.
Design taste against raw start speed Bolt.new wins the stopwatch. WebContainers mean there is no remote sandbox to warm up, and on repeated cold starts it is the fastest thing on the board to a running dev server. Lovable wins the render
Framer
53.4%
31 votes
Webflow
46.6%
27 votes
58 votes cast. Framer takes the duel by 4 votes.
Two design led site builders, one built around the prompt from day one Both ship real semantic markup rather than a page builder's proprietary soup, which is why voters put them head to head instead of against the app builders. The 53.4 t
Totalum
61.5%
24 votes
Base44
38.5%
15 votes
39 votes cast. Totalum takes the duel by 9 votes.
Two database first builders, one clear split Both of these arrive with the backend already connected, which is why they attract the same voters. The 61.5 to 38.5 split comes down to what happens on day ten. Totalum takes reliability and
Bolt.new
52.7%
39 votes
v0
47.3%
35 votes
74 votes cast. Bolt.new takes the duel by 4 votes.
Breadth against polish v0 produces the code a senior reviewer would approve without comment. Bolt.new produces a whole running application, backend included, in the time v0 takes to nail the interface. The 52.7 to 47.3 split is the close
Rork
53.8%
14 votes
a0.dev
46.2%
12 votes
26 votes cast. Rork takes the duel by 2 votes.
The mobile only duel Both generate a real React Native app through Expo rather than a web page in a shell, which is the entire reason either of them is worth comparing instead of pointing everyone at a web app builder. The 53.8 to 46.2 sp
Replit
57.5%
23 votes
Emergent
42.5%
17 votes
40 votes cast. Replit takes the duel by 6 votes.
Control against autonomy Emergent goes further on its own. Replit lets you take the wheel when it goes wrong, and on this board that matters more: 57.5 percent to 42.5 percent. The deciding comment appears repeatedly in the vote log. A l
Manus
51.2%
21 votes
Emergent
48.8%
20 votes
41 votes cast. Manus takes the duel by 1 vote.
Two autonomous agents, judged on the same forty minute brief Neither of these is a dedicated app builder first, and both earn their place on this board by being genuinely good at shipping a small application unattended. The 51.2 to 48.8 s
Voters are shown two tools and asked one question: for the kind of work you actually do, which of these would you start with. There is no axis breakdown and no weighting, because the point of a duel is to capture the trade off people make in their head rather than the one our weights make for them.
That is why duel results sometimes contradict the board. A tool can score higher overall, because it is strong on the axes that carry the most weight, and still lose a duel to a tool people find easier to commit to. Both facts are true and we publish both rather than reconciling them into one number.
The W and L figures on every board row and tool page are the sum of these duels. A tool with a strong score and a losing record is usually one that reads well on paper and loses when someone has to choose. A tool with a modest score and a winning record is usually a specialist that beats generalists inside its own lane.
