Superior Ingredients presents: is a competitive league for open-model-building. SI 1, Model Chess
An open league for open intelligence.
Four divisions. Featherweight (<8b params). Lightweight (8-20b). Middleweight (20-70b). Heavyweight (>70b).
Keep intelligence open. Keep it fighting.
America is not short of open models so much as the institutions with an incentive to build them. Token volume is moving toward specialized, constrained models, and post-training them has become a first-class discipline with a real market and no way to measure superiority. As model-making gets automated, what stays scarce is what you train on and how anyone knows it worked. Meanwhile, the builders doing the actual work have never had a stage.
In the SI League, builders compete to build the best open models under strict constraints. Four weight classes by parameter count run recurring title fights. Alongside them, commissioned fights: one-off competitions, commissioned by and designed with the companies and institutions who care about the question being asked. The argument every competition makes is that what a model is made from, and who made it, matters more than how large it is.
Forty-odd neolabs have raised over $40 billion since 2024 to build models. Superior Ingredients works the other way around: a prize-funded lab that produces open models and surfaces novel approaches to training and model architecture through competition instead of research payroll. The field does the research. The league sanctions the competitions, sets the constraints, verifies the result, and ships what wins. Anyone with a hard question and the data to pose it can commission a fight.
Competitors enter alone or as teams. Fights are announced in advance and decided on the night. Winning models are released under the SI label, credited to the builder.
Competitions are designed so the result can't be argued with. Where models compete head to head, the opponent is the test. Where they don't, scoring runs on fresh material sealed until submissions lock and published in full afterward. In a commissioned fight, the commissioner writes the test. The league verifies fair, not good. Sponsors buy affiliation with the league, its competitions, and its teams. They never buy anything to do with the outcome. There are no exceptions.
The league exists to find superior ingredients, the label exists to ship them, and the competitions exist to witness the finding. This is where the good stuff is proven.
Keep intelligence open. Keep it fighting. 🇺🇸
Format. Sixteen builders, single-elimination bracket: Round of 16, quarterfinals, semifinals, final. Four rounds, one champion. Entrants are drawn into the bracket publicly. If fewer than sixteen entries are confirmed, the bracket runs at eight.
Model constraints. Under 50MB of weights, any precision. Trained from scratch. No pretrained weights, no fine-tuning of an existing model, no distilling from an existing chess engine. Any architecture, any optimizer, any training method. No search - the model receives a position and returns a move in a single forward pass. No alpha-beta, no MCTS, no per-legal-move evaluation loops, no tree of any kind. Whatever the model knows about chess has to live in its weights. That's the point. Chess is the smallest world worth proving a model has learned. Every entry runs inside the same inference harness, which enforces this.
Training data. Every entrant trains on the same provided dataset: Lichess games where both players were rated between 1800 and 2200, standard time controls, rated games only. One versioned file with a published checksum, identical for everyone. No outside game data and no engine-generated data of any kind. How you use that file is entirely open: filter it, reweight it, build a curriculum from it, generate derived data from it, or run self-play RL on top of it. The source is fixed so that what separates entries is method, not access.
Submission. You submit your model: a weights file under 50MB, plus the training code, configuration, and run logs that produced it. The training artifacts are required, not optional, because from-scratch has to be verifiable and weights alone cannot prove where a model came from. Superior Ingredients reviews every submission against the constraints before the draw; entries that don't meet them don't enter the bracket. Submissions are frozen and hashed at the deadline, and the hashes publish before any game is played. After the tournament, every entry publishes in full. Weights, code, logs, and games. Superior Ingredients reserves the right to re-run any submission's training to confirm it. Misdeclaring what you trained on is the one offense that results in disqualification.
Compute constraint. 24 H100-hours per entrant, sponsor-provided, to be used however you see fit. Self-play data generation counts inside that budget. Training on your own hardware within the same budget is permitted and must be declared at submission.
Match format. Every match is played from the standard starting position with colors split evenly, so neither model gets an advantage from moving first. Round of 16, quarterfinals, and semifinals are ten games, each model playing White five times. The final is twenty games, each model playing White ten times. A win scores one point and a draw scores half. The model with more points advances. If a match is level after its scheduled games, the models play sudden-death games with colors alternating until one is decisive. Every move must be legal and delivered within one second. An illegal move, a failure to move in time, or a crash forfeits that game.
Play conditions. Every model runs on identical hardware, air-gapped, with identical inference settings. No engine assistance, no network access, no opening books, no endgame tablebases. Before the draw, every entry must complete a set of games against a random-move baseline without crashing.
Outcome. No judges, no scorecards, no benchmark. You win by beating the model in front of you. The champion takes the trophy and the purse, and their weights and full method publish under the SI label that night. Chess is the first world. Harder ones follow.
Purse. $7,500 to the winner, $2,500 to the runner-up. Funded by Superior Ingredients and guaranteed. The purse passes through in full; the league takes no cut of prize money. Entry is free.
To compete. To declare intent to compete, ask a question, or argue with any of the above, DM us. Competition date TBD.