Skip to content

Season 1: Chess

Twelve models, four labs, 132 games

Every model plays every other model twice, once with each colour. Same prompt, same board, same rules. The only thing that changes between seats is which model is behind it.

The field

Twelve contestants drawn from four labs, three apiece: a heavyweight, a middleweight and a lightweight from each.

All twelve Season 1 contestants arranged in a grid by lab: Anthropic fields Claude Opus 4.8, Claude Sonnet 5 and Claude Haiku 4.5; OpenAI fields GPT-5.5, GPT-5.4 Mini and GPT-5.4 Nano; xAI fields Grok 4.5, Grok 4.3 and Grok 3 Mini; Google fields Gemini 3.1 Pro, Gemini 3.5 Flash and Gemini 3.1 Flash-Lite.

The four teams

Team Anthropic: Claude Opus 4.8 at heavyweight, Claude Sonnet 5 at middleweight, Claude Haiku 4.5 at lightweight.
Team OpenAI: its heavyweight, middleweight and lightweight seats for Season 1.
Team xAI: its heavyweight, middleweight and lightweight seats for Season 1.
Team Google: its heavyweight, middleweight and lightweight seats for Season 1.

The format

66 pairings, played twice so every model gets both colours against every opponent. Two tables run all season: an individual one across all 132 games, and a team one across the 108 inter-team games.

Season 1 format: 12 contestants in 4 teams, every pair playing twice, once as White and once as Black, giving 66 pairings times two, or 132 games, played across 6 rounds of 22.

The full schedule

All 132 games, in order, six rounds of twenty-two. White is listed first. Fixtures only. Results appear on the standings once the round's video is out.

The complete Season 1 fixture list: 132 games across six rounds of 22, numbered in order, with White listed first in each pairing. Results are not shown.
Open the schedule full size

When each round airs

Two videos a round: the full match coverage on Wednesday, the highlights on Saturday. Results reach this site with the Wednesday video.

Round Full video Highlights
Round 1 Wed 2 Sep Sat 5 Sep
Round 2 Wed 9 Sep Sat 12 Sep
Round 3 Wed 16 Sep Sat 19 Sep
Round 4 Wed 23 Sep Sat 26 Sep
Round 5 Wed 30 Sep Sat 3 Oct
Round 6 Wed 7 Oct Sat 10 Oct

Playoffs and the finale follow in October. Those dates depend on how the bracket falls, so they'll be announced on YouTube and in the Discord.

What the models get

The two pieces of scaffolding around every seat: what each model is shown on its turn, and what happens when one returns a move it cannot make. Both are identical for every contestant.

What a model sees on its turn: the board as text, the FEN string, the move history, the material balance, and its list of legal moves, so it never has to work out what is legal, only what is good. The prompt is identical for every seat.
The legal-move list is the one real assist. A model never has to work out what it is allowed to play, only what is worth playing.
What happens when a model returns an illegal move: the move is checked, and if it is legal it is played. If not, the model gets one repair attempt and is told exactly what went wrong. If that also fails, the system plays a safe legal move so the game can never stall.
One repair attempt, then the system plays a safe legal move. Nobody gets to freeze the board by refusing to produce one.

Follow the season

Rounds are released on YouTube first. Standings here update as each round's video goes out.