Sequoia Capital, OpenAI and Citadel Securities have opened applications for No Limit, a pokerbot-building tournament offering a $100,000 grand prize. Applications opened October 27, 2026, with three November regionals leading to a November 15 final in San Francisco. Teams of one or two get an initial 60 minutes to build their bots. The tournament page advertises 10,000 simulated games against other entrants.
This is a programming competition, not an OpenAI model release. Coding assistants are allowed during development, but submitted bots must play without network access and return each decision within one millisecond.
The official rules identify Sequoia Capital Operations as the organizer and OpenAI and Citadel Securities as hosts. Sequoia handles administration, eligibility checks, scoring, winner selection and prize payment. For applicants, the consequential details are the attendance requirements, the short registration window and the gap between using AI to write code and running that code during play.
Who Can Enter No Limit?
Entry is free and open to:
- Currently enrolled students at universities in the United States.
- Graduates of U.S. universities from the classes of 2025 or 2026.
- Participants aged 18 or older who meet the rules’ legal eligibility requirements.
Teams can have one or two people, and teammates do not need to attend the same school. Each participant may compete in only one regional. No purchase of OpenAI services or another product is required. Employees and contractors of the competition organizations, along with specified family and household members, are excluded under the rules.
Both teammates must register individually. The application emails the second teammate a form, and the team becomes complete when that person submits it. Current students are asked to use their school email for enrollment verification; recent graduates may use any email.
Places are limited and offered on a rolling basis. I’d apply before the deadline if the travel dates work: rolling acceptance means the closing date is not a guarantee that places will remain available.
Regional Deadlines, Travel and the $100,000 Prize
All events below take place in 2026. The public page uses Bay Area and Boston as regional labels; the rules specify Menlo Park and Cambridge as the locations.
| Event | Date | Registration Deadline |
|---|---|---|
| Bay Area: Menlo Park, California | Friday, November 6 | November 3, 11:59 p.m. PT |
| Boston area: Cambridge, Massachusetts | Monday, November 9 | November 6, 11:59 p.m. ET |
| Pittsburgh, Pennsylvania | Thursday, November 12 | November 9, 11:59 p.m. ET |
| Finals: San Francisco, California | Sunday, November 15 | Invitation only, based on regional performance |
Regional check-in begins at 5:30 p.m., and each event runs 6–9 p.m. local time. Participants must attend the full regional in person. The San Francisco final runs noon–8 p.m., also in person. These dates and requirements appear in the rules and application.
Top-performing regional teams receive invitations to the final. The landing page says they will be flown to San Francisco, and the application states that travel is covered. Participants must be able to attend the final to win the final prize.
The advertised $100,000 is a grand prize for the winning team, not a stated per-person award. The available source material does not provide a complete payout breakdown or itemized travel terms. Finalists should confirm covered expenses, booking arrangements and reimbursement procedures before making purchases.
The schedule leaves little room for rearranging commitments. Pittsburgh competitors, for example, have only three calendar days between their regional and the final. Checking November 15 availability before applying is a practical necessity.
The 60-Minute Build Is Only the Opening Phase
Teams receive a starter kit containing:
- A Python agent template and provided
Agentinterface. - A local simulator.
- An Independent Chip Model, or ICM, calculator.
- The payout structure and blind schedule.
An ICM calculator estimates the prize-value implications of chip stacks under a payout structure. Gaining chips and maximizing expected prize money can call for different decisions, so the distinction matters.
The rules permit Astra and Codex to help teams write, debug and refine their bots during the build phase. Publicly available documentation is also allowed. Teams may submit one bot every five minutes, with the most recently accepted submission becoming the active entry.
The first 60 minutes are not the entire development window. The rules specify six scored tournaments, labeled T1–T6. The first starts when the initial build phase ends, and subsequent tournaments start 20 minutes apart. Teams may continue editing and submitting bots after that opening hour.
That changes the sensible development strategy. A working baseline has immediate value because it can compete while a team improves its decision logic. Spending the whole opening hour on an ambitious approach that never produces a valid submission would leave little protection against implementation mistakes.
The Exact Poker Variant Is Revealed at the Event
The landing page promotes 10,000 simulated no-limit hold’em games, but the official rules say the poker variant remains confidential until each event’s build phase begins.
Applicants should distinguish the promotional description from the operative rule. The published rules leave the exact variant undisclosed until the clock starts, so preparing exclusively for one fixed hold’em setup would carry a risk.
The useful preparation is broader: understand the supplied interface when it arrives, inspect the simulator and payout structure, and build a valid decision loop quickly. That recommendation follows from the documented uncertainty about the variant, not from any firsthand testing of the platform.
Offline Play Makes Reliability Part of the Strategy
The runtime restrictions sharply limit what an entrant can submit. According to the official rules:
- No network access: Bots run in isolated sandboxes and cannot call OpenAI tools, external APIs or outside services for gameplay decisions.
- One-millisecond decisions: Each bot must return a move within 1 ms, or one-thousandth of a second.
- Automatic fallback: An error or timeout causes the bot to fold, or check when checking is available.
- Limited persistence: Bot state resets at the start of each table.
- Anonymous opponents: Bots cannot identify specific competing teams.
The rules also specify a maximum file size of 100mb.
These conditions make AI assistance a development resource, while the submitted artifact must function as a fast, self-contained decision-making program. A team can use a coding assistant to help construct its strategy, but it cannot send a hand to a remote LLM and wait for advice during play.
I’d put predictable execution ahead of elaborate decision logic here. The reason is concrete: missing the time limit directly changes the bot’s action to a check or fold. A sophisticated policy that frequently times out gives up control precisely when it needs to make a decision.
State resets and opponent anonymity also constrain adaptation. Teams should not assume they can maintain a persistent profile of a named rival across tables.
Only bots play in the competition. Chips and points have no cash value, no real money is bet by the bots, and attendees may not wager on outcomes. The cash award is a competition prize, separate from the simulated poker stakes.
Frequently Asked Questions
3 questions
1Can I Use OpenAI Tools to Build My Bot?
Yes. The rules allow Astra and Codex during development. Submitted bots must play offline without calls to OpenAI tools or other external services.
