CrowdSota Open
Public science, aimed at published SOTA
Anyone can join the public lab. Contributions go on a public ledger, with credit that can become authorship. The aim is crowd-sourced research that reaches published SOTA.
Join the open waitlistOne core · two front doors
CrowdSota is one core: a fair-credit contribution ledger plus an AI research loop, where people contribute tokens, compute, or ideas. That same core powers two products of equal weight.
CrowdSota Open
Anyone can join the public lab. Contributions go on a public ledger, with credit that can become authorship. The aim is crowd-sourced research that reaches published SOTA.
Join the open waitlistCrowdSota for Enterprise
Leadership sets strategic goals and attaches budgets. Anyone, across team lines, can contribute at any time. Fair, visible credit — and the strongest work can be funded as real products.
Explore the enterprise doorFundraising site · Launch target 22 November 2026 · Product not yet built · View the pitch deck
The problem
The same gap shows up in public research and inside companies: people and agents can already do the work, but credit, access, and funding still sit in closed rooms, or in a once-a-year burst.
In open science
AI agents can already search literature, run experiments, and draft papers. The tools often live in a private workspace. Decentralized networks may reward GPUs or tokens, which is not authorship. Open-science platforms welcome people, but are not built for bring-your-own keys or agents on a public ledger.
Inside companies
Internal hackathons are either lightweight individual exercises that do not take on large problems, or serious annual events that are effort-heavy and then go dark. Teams work in silos. There is often no direct, ongoing incentive, and leadership cannot see how AI spend maps to strategic goals — across, for example, a company with 100 teams of about five people working independently.
How it works
The loop is the same in CrowdSota Open and CrowdSota for Enterprise. People contribute tokens, compute, or ideas; a fair-credit ledger records the work. In Open, that ledger is public. In Enterprise, it is visible across team lines, inside the company.
Tokens
Use your own model API key for a single run (BYOK), or donate credits to a shared pool that others can draw from. The ledger records which path funded the work.
Compute
A lightweight worker picks up queued experiments. Samples are spot-checked by re-running them elsewhere, so a contribution is not just a claim — it can be checked.
Ideas, code, reviews
Propose a question, ship code, or review a result. Human judgment sits beside agent work, and both are attributed.
Fair credit
Credit is not a vibe and it is not only a token balance. We separate measurable effort from evidence of impact, and we map people (and agents) onto standard contributor roles when a result is written up. In Open, the ledger is public. In Enterprise, the same rules make credit visible across team lines.
Tokens spent, GPU-hours contributed, runs completed. This layer answers: who showed up, and with what resources?
Runs that move a stated benchmark score count more than runs that do not. This layer answers: whose work changed the result?
When a project is written up, authorship follows the CRediT taxonomy (an ANSI/NISO standard): roles such as Conceptualization, Software, Validation, and Writing — not a single undifferentiated byline.
Example — illustrative only, not live data
| Contributor | Type | Effort | Impact | CRediT role |
|---|---|---|---|---|
| A. Chen | ideas | Hypothesis + protocol | Accepted into challenge #1 | Conceptualization |
| @gpu-node-14 | compute | 18.4 GPU-hours, 12 runs | 3 runs improved the score | Investigation |
| R. Okonkwo | tokens | BYOK, 4 agent reviews | Caught a failed spot-check | Validation |
| credit-pool | tokens | Shared credits, 2 runs | No score move | Resources |
Names and figures in the table are fictional examples to show how the ledger would read. They are not traction.
CrowdSota Open
Anyone can participate. Contributions of tokens, compute, and ideas go on a public ledger, with fair credit that can become authorship. The long-term aim is crowd-sourced research that reaches published state of the art — we will not claim a SOTA result until one exists.
01
You do not need to be at a frontier lab. Researchers, contributors, and sponsors can join a public challenge and be named for the work they actually did.
02
Effort (tokens, GPU-hours, runs) and impact (runs that move a benchmark score) are both visible. When a project is written up, authorship follows CRediT roles.
03
The point is not a token balance. It is a shared, checkable record of research that can be reviewed, reproduced, and, if it earns it, published.
CrowdSota for Enterprise
Leadership defines big strategic goals and attaches budgets — tokens, compute, funding, and recognition. Any employee or team, across organizational silos, can contribute ideas, experiments, code, or compute at any time. The ledger gives fair, visible credit across team lines. The strongest contributions can be funded to become real products.
Goals & budgets
Strategic goals stay in view all year, not only during a hackathon week. Budgets travel with the goal, so people can keep making progress when the idea is ready — not when the calendar says so.
Across silos
A contribution can come from any team, not only the team that “owns” the problem. Credit is visible across team lines, so work is not trapped in a local hero’s weekend project.
Incentives
The ledger is the incentive: effort and impact are named. Top contributions and projects can be funded for implementation as products — continuously, not as a once-a-year prize.
For leadership
Because tokens, compute, and outcomes sit on the same ledger as the stated goals, leadership gets a live view of where AI spend is actually going — and what it is moving.
From the founder — workplace experience, not a customer claim
Internal hackathons were fine for people to join as individuals. They were weekend-scale, “toy” exercises. They were not a way to solve large problems.
The annual hackathon is serious: the top five products get funding to be implemented as products. It still happens once a year, and standing it up takes a lot of effort.
CrowdSota is meant to automate that into a continuous, year-long hackathon, with direct incentives, so anyone in the company can make progress at any time. It is also meant to break silos — for example, in a company with 100 teams of about five people working independently today. Amazon and Toyota are mentioned here only as places the founder has worked, not as customers or design partners.
Enterprise roadmap proposed
The first enterprise version is planned to add controls that a public lab does not need. These are roadmap items, not shipping features.
Why now
The research observations below are from sources we reviewed in November 2026, not market-size claims. The enterprise framing is founder experience, not a survey.
How we’re different
For CrowdSota Open: open participation, BYOK plus a shared credit pool, agents that do research work, and a public fair-credit ledger. Adjacent products cover some of these, not all. This comparison is from a November 2026 review of public pages, not a paid landscape study.
| CrowdSota planned | Meshia / OpenScience | Sakana / FutureHouse / Autoresearch / Orikata | Bittensor / Prime Intellect | ResearchHub / Zooniverse | |
|---|---|---|---|---|---|
| Open participation | Yes — public lab | Private team workspaces | Single-team systems | Open for compute / mining | Open to people |
| BYOK + shared credits | Yes — key per run or pool | BYOK / wallet / credits in-product | Typically hosted or single-team | Compute and token rewards | Not built around model keys |
| AI agents doing research | Yes — queued experiments | Yes | Yes | Training / inference, not paper authorship | Human review or classification |
| Provable fair credit | Public ledger + CRediT | No public contribution ledger found | Ordinary internal or paper credit | Emissions for work, not CRediT authorship | Open, but not agent + BYOK credit |
For CrowdSota for Enterprise, the contrast is not that public table. It is the difference between a weekend or annual hackathon and a continuous, funded challenge with a ledger that crosses team lines. We are not naming other internal-innovation vendors here; we have not done a paid landscape study of that category.
Roadmap
22 November 2026
A public pitch, a pledge form, and conversations with researchers, contributors, sponsors, investors, and companies interested in a pilot. The product is not live yet.
Next proposed
One narrow, measurable public task. Waitlist pledges of keys, GPU-hours, and ideas are invited to join.
Then proposed
A visible log of tokens, GPU-hours, and runs — including BYOK versus pool. GPU workers pull queued experiments. A sample of runs is re-executed to catch bad results.
In parallel proposed
An always-on internal challenge tied to a leadership goal and a budget, with credit visible across team lines. Top work can be put forward for product funding.
Enterprise needs proposed
Single sign-on, private deployment and data controls, and an audit trail — the controls companies will need before this runs on internal work.
Aim proposed
Score moves feed the impact layer. Open results can be authored with CRediT roles; the long-term aim is published SOTA, claimed only if it exists. Inside a company, the corresponding aim is that the best work is funded as a real product.
Pitch deck
Use the arrows, the keys ← →, or fullscreen. Placeholders for the raise and the founding team are marked in gold. Open and Enterprise have equal weight.
Waitlist & pledges
Tell us who you are and what you might contribute, or if you want to talk about an enterprise pilot. This is a pledge of interest, not a contract and not a sale of tokens or equity.
If you want to talk about the raise, use the button. It opens the waitlist form with your role set to Investor.