short answer
As of Tue, 17 Nov 2026, the TensorHack winners library covers 4,426 winning projects from 334 hackathons on Devpost, ETHGlobal and Devfolio. 76% of winning projects link a public code repository and 84% link a demo video; 65% have both. Among 3,709 winners that list their team, the most common size is one person (38%), and 38% won solo.
on this page
key findings
- As of Tue, 17 Nov 2026, the TensorHack winners library covers 4,426 winning projects from 334 hackathons on Devpost, ETHGlobal and Devfolio.
- 76% of winning projects link a public code repository and 84% link a demo video; 65% have both.
- Among 3,709 winners that list their team, the most common size is one person (38%), and 38% won solo.
- AI is the most common theme, on 58% of winning projects, followed by Web3 (22%) and Gaming (10%). A project can carry several themes.
- 13% of winning projects took more than one prize or track award.
code and demo video
What winning projects link
As of Tue, 17 Nov 2026. 4,426 winning projects
| What winning projects link | winning projects | share |
|---|---|---|
| a public code repo | 76% | |
| a demo video | 84% | |
| both | 65% | |
| neither | 4% |
The links are what each project page shows. A missing link can mean the team kept the code private or never added it to the page, not that the project had none.
team size
People on winning teams
As of Tue, 17 Nov 2026. 3,709 winners that list their team
| People on winning teams | winning projects | share |
|---|---|---|
| solo | 38% | |
| 2 people | 20% | |
| 3 people | 16% | |
| 4 people | 25% | |
| 5 or more | 1% |
themes
Themes of winning projects
As of Tue, 17 Nov 2026. a project can carry several themes; themes under 5 projects are left out
| Themes of winning projects | winning projects | share |
|---|---|---|
| AI | 58% | |
| Web3 | 22% | |
| Gaming | 10% | |
| Social good | 8% | |
| Dev tools | 6% | |
| Fintech | 6% | |
| Education | 5% | |
| Health | 5% | |
| Hardware | 4% | |
| Security | 4% | |
| Climate | 3% |
prizes per project
Prizes or track awards per winning project
As of Tue, 17 Nov 2026. 4,426 winners with listed prizes
| Prizes or track awards per winning project | winning projects | share |
|---|---|---|
| one prize | 87% | |
| two prizes | 10% | |
| three or more | 2% |
platforms and hackathons
Where the winners were listed
As of Tue, 17 Nov 2026.
| Where the winners were listed | winning projects | share |
|---|---|---|
| Devpost | 77% | |
| ETHGlobal | 14% | |
| Devfolio | 9% |
Hackathons with the most winning projects in the library (3 or more):
- ForgeHacks Online 2026: 200 winning projects
- Frostbyte Hackathon: 186 winning projects
- HackMoney 2026: 166 winning projects
- Amazon Nova AI Hackathon: 104 winning projects
- ETHGlobal New York 2026: 99 winning projects
- ETHGlobal Mumbai: 74 winning projects
- ETHDenver 2026: 73 winning projects
- ETHGlobal Cannes 2026: 67 winning projects
- World’s Largest Hackathon presented by Bolt: 64 winning projects
- ETHOnline 2026: 63 winning projects
This reflects which galleries the library has read so far, not which hackathons are best. Browse every project on the winners library.
How to read these numbers
Every figure on this page is counted from the TensorHack winners library, which collects winning projects from the public project galleries on Devpost, Devfolio and ETHGlobal and links each one back to its source page. The library refreshes once a day.
The library holds winners only. A share like "most winners carry the AI theme" describes what winners look like. It does not say AI projects win more often, because the projects that did not win are not in the data: if most entries at an event were AI projects, most of its winners would be too.
A split appears only when at least 30 winning projects carry the field it counts, and a theme or hackathon is listed only with at least 5 or 3 projects. Smaller groups are left out rather than turned into shaky percentages.
What the data cannot tell you
- Judging scores. Galleries show who won, not how judges scored the entries.
- Languages and frameworks. Built-with tags are used only to assign themes, so this page does not rank tech stacks.
- Everyone on the team. Team lists show members with a public profile on the platform; a teammate without one is not counted.
- Prize size. Prize names vary by host, from "First place" to "Best use of" a sponsor's API, so prizes are counted, not valued.
How to use it at your next hackathon
Treat the common traits of winners as the floor your submission should clear, then read the event's own rules, which always come first. The judging criteria guide shows how rubrics score a project.
- If most winners link code, keep a public repo with a clear README from the first hour. The Devpost submission guide covers the fields.
- If many winners link a demo video, record it before the deadline rush. The demo guide covers length and structure.
- For team size, compare the numbers with how to build a hackathon team and solo vs team hackathons.
For the full playbook, read how to win a hackathon, then browse the winners library for projects in your theme.
FAQ
Do hackathon winners usually have a GitHub repo?
In the TensorHack winners library, 76% of 4,426 winning projects link a public code repository and 84% link a demo video (as of Tue, 17 Nov 2026).
How big are winning hackathon teams?
Among 3,709 winners that list their team, one person is the most common size (38%), and 38% won solo. Team lists only show members with a public profile.
What kind of projects win hackathons?
AI is the most common theme among winning projects in the library, on 58% of them. The library holds winners only, so this shows what winners most often build, not which themes have better odds.
methodology and citation
Counts come from the winners library: winning projects read once a day from public galleries on Devpost, Devfolio and ETHGlobal, stored as short facts with a link to each source page. Themes are assigned from the hackathon's own categories, the project's built-with tags and its tagline. Team size is the number of members the project page lists. A project counts as having code or a video when its page links one. Nothing is estimated, and splits with fewer than 30 rows are left out. There is no data download.
The figures change daily, so cite the date you read them:
TensorHack. (2026). What Actually Wins Hackathons: Winner Data (figures as of Tue, 17 Nov 2026). https://tensorhack.com/guides/what-wins-hackathons