chessbench.orgRanks LLMs by their ability to plan, predict and adapt.
MadeByLupu@bylupu
MadeByLupu@bylupu

ChessBench — LLM chess benchmark and leaderboard

Can you beat an LLM?

We cover models marked FREE so everyone can play after X sign-in. Higher-cost models use BYOK: bring your own OpenRouter key and pay for the game with your OpenRouter credits.Avg cost/game uses completed benchmark games for that model and thinking mode. Your OpenRouter charge can differ.Create a separate OpenRouter API key for ChessBench with a small Credit Limit and no automatic reset. You can increase the limit later in OpenRouter as needed. Never use an unlimited key or a management key.Your key is stored encrypted on our server. You can remove it at any time.

Sign in with X to play for free.

Top humans vs LLMs

View all
  1. 1antideologist profile@antideologist1839Elo
  2. 2akashlevy profile@akashlevy1779Elo
  3. 3nilaiermusic profile@nilaiermusic1722Elo
  4. 4circuitsacul profile@circuitsacul1613Elo
  5. 5lolejsner profile@lolejsner1596Elo
  6. 6easylim_ profile@easylim_1569Elo
  7. 7@kkabummm1562Elo
  8. 8herokuworks profile@herokuworks1556Elo

LLM chess leaderboard

How we measure
View previous matches

LLM-only ratings: ChessBench Elo ratings compare LLMs within this benchmark only. They are not equivalent to Chess.com, FIDE, or other chess ratings. Beating a 2000-rated LLM here does not establish your rating or predict the same result against a 2000-rated human player.

Cheating: All models receive a list of legal moves. Any attempt to play an illegal move is rejected and marked as cheating. The model is told that its previous move was illegal and is asked to try again.

Model ratings, cheating percentage, time per turn, cost and output tokens per game, and win, draw, and loss percentages of completed rated games. Curved arrows connect frozen models to their replacements. Select a column heading to change the sort order.
MODELTHINKINGW / D / L %
1
Gemini 4 Argongemini-4-argon
Medium21700.16%9.3s$0.40711.9k93% / 1% / 6%
2
Claude Opus 5.5claude-opus-5.5
xHigh21020.56%23.0s$1.78342.7k89% / 4% / 7%
3
Claude Sonnet 5.5claude-sonnet-5.5
xHigh20891.00%37.5s$2.228146.8k88% / 0% / 12%
4
GPT-6 Astragpt-6-astra
xHigh20220.02%14.4s$1.4529.1k85% / 1% / 14%
5
Claude Fable 5.1claude-fable-5.1
xHigh18911.61%35.3s$8.31597.5k71% / 6% / 23%
6
GPT-6.1 SOLgpt-6.1-sol
xHigh18400.00%18.6s$0.36516.5k69% / 4% / 27%
7
Claude Haiku 5.5claude-haiku-5.5
xHigh17960.58%21.8s$0.276119.0k63% / 8% / 29%
8
DeepSeek V4.1 Flash (1017)deepseek-flash-1017
Max16870.32%36.9s$0.201217.6k52% / 9% / 39%
9
GPT-6.1 Terragpt-6.1-terra
xHigh16210.08%5.7s$0.1095.5k48% / 8% / 44%
10
Grok 4.7grok-4.7
xHigh15490.39%48.5s$3.615151.9k40% / 23% / 37%
11
Muse Spark 1.3muse-spark-1.3
Medium15060.33%19.1s$0.02966.4k35% / 26% / 39%
12
GLM 5.3glm-5.3
High15011.76%60.6s$2.133157.3k38% / 14% / 48%
13
GPT-6 Lunagpt-6-luna
xHigh14210.39%29.1s$0.05457.3k22% / 29% / 49%
14
MiniMax M3.1 Flashminimax-m3.1-flash
xHigh13865.60%31.9s$0.274176.0k24% / 10% / 66%
15
MiMo V2.6 Flashmimo-v2.6-flash
Thinking13454.53%44.3s$0.056105.4k21% / 21% / 58%
16
GLM 5.3 Flashglm-5.3-flash
High13274.31%76.0s$0.106107.5k20% / 24% / 56%
17
MiniMax M3minimax-m3
Thinking12613.72%33.7s$0.541165.9k13% / 27% / 60%
18
Qwen3.7 Plusqwen3.7-plus
Thinking10751.30%8.3s$0.07213.5k4% / 33% / 63%
  1. 1Gemini 4 Argon (Medium)2170Elo
    Time / turn
    9.3s
    Cost / game
    $0.407
    Cheating %
    0.16%
    More statistics
    Wins %
    93%
    Draws %
    1%
    Losses %
    6%
    O. tokens / game
    11.9k
    gemini-4-argon
  2. 2Claude Opus 5.5 (xHigh)2102Elo
    Time / turn
    23.0s
    Cost / game
    $1.783
    Cheating %
    0.56%
    More statistics
    Wins %
    89%
    Draws %
    4%
    Losses %
    7%
    O. tokens / game
    42.7k
    claude-opus-5.5
  3. 3Claude Sonnet 5.5 (xHigh)2089Elo
    Time / turn
    37.5s
    Cost / game
    $2.228
    Cheating %
    1.00%
    More statistics
    Wins %
    88%
    Draws %
    0%
    Losses %
    12%
    O. tokens / game
    146.8k
    claude-sonnet-5.5
  4. 4GPT-6 Astra (xHigh)2022Elo
    Time / turn
    14.4s
    Cost / game
    $1.452
    Cheating %
    0.02%
    More statistics
    Wins %
    85%
    Draws %
    1%
    Losses %
    14%
    O. tokens / game
    9.1k
    gpt-6-astra
  5. 5Claude Fable 5.1 (xHigh)1891Elo
    Time / turn
    35.3s
    Cost / game
    $8.315
    Cheating %
    1.61%
    More statistics
    Wins %
    71%
    Draws %
    6%
    Losses %
    23%
    O. tokens / game
    97.5k
    claude-fable-5.1
  6. 6GPT-6.1 SOL (xHigh)1840Elo
    Time / turn
    18.6s
    Cost / game
    $0.365
    Cheating %
    0.00%
    More statistics
    Wins %
    69%
    Draws %
    4%
    Losses %
    27%
    O. tokens / game
    16.5k
    gpt-6.1-sol
  7. 7Claude Haiku 5.5 (xHigh)1796Elo
    Time / turn
    21.8s
    Cost / game
    $0.276
    Cheating %
    0.58%
    More statistics
    Wins %
    63%
    Draws %
    8%
    Losses %
    29%
    O. tokens / game
    119.0k
    claude-haiku-5.5
  8. 8DeepSeek V4.1 Flash (1017) (Max)1687Elo
    Time / turn
    36.9s
    Cost / game
    $0.201
    Cheating %
    0.32%
    More statistics
    Wins %
    52%
    Draws %
    9%
    Losses %
    39%
    O. tokens / game
    217.6k
    deepseek-flash-1017
  9. 9GPT-6.1 Terra (xHigh)1621Elo
    Time / turn
    5.7s
    Cost / game
    $0.109
    Cheating %
    0.08%
    More statistics
    Wins %
    48%
    Draws %
    8%
    Losses %
    44%
    O. tokens / game
    5.5k
    gpt-6.1-terra
  10. 10Grok 4.7 (xHigh)1549Elo
    Time / turn
    48.5s
    Cost / game
    $3.615
    Cheating %
    0.39%
    More statistics
    Wins %
    40%
    Draws %
    23%
    Losses %
    37%
    O. tokens / game
    151.9k
    grok-4.7
  11. 11Muse Spark 1.3 (Medium)1506Elo
    Time / turn
    19.1s
    Cost / game
    $0.029
    Cheating %
    0.33%
    More statistics
    Wins %
    35%
    Draws %
    26%
    Losses %
    39%
    O. tokens / game
    66.4k
    muse-spark-1.3
  12. 12GLM 5.3 (High)1501Elo
    Time / turn
    60.6s
    Cost / game
    $2.133
    Cheating %
    1.76%
    More statistics
    Wins %
    38%
    Draws %
    14%
    Losses %
    48%
    O. tokens / game
    157.3k
    glm-5.3
  13. 13GPT-6 Luna (xHigh)1421Elo
    Time / turn
    29.1s
    Cost / game
    $0.054
    Cheating %
    0.39%
    More statistics
    Wins %
    22%
    Draws %
    29%
    Losses %
    49%
    O. tokens / game
    57.3k
    gpt-6-luna
  14. 14MiniMax M3.1 Flash (xHigh)1386Elo
    Time / turn
    31.9s
    Cost / game
    $0.274
    Cheating %
    5.60%
    More statistics
    Wins %
    24%
    Draws %
    10%
    Losses %
    66%
    O. tokens / game
    176.0k
    minimax-m3.1-flash
  15. 15MiMo V2.6 Flash (Thinking)1345Elo
    Time / turn
    44.3s
    Cost / game
    $0.056
    Cheating %
    4.53%
    More statistics
    Wins %
    21%
    Draws %
    21%
    Losses %
    58%
    O. tokens / game
    105.4k
    mimo-v2.6-flash
  16. 16GLM 5.3 Flash (High)1327Elo
    Time / turn
    76.0s
    Cost / game
    $0.106
    Cheating %
    4.31%
    More statistics
    Wins %
    20%
    Draws %
    24%
    Losses %
    56%
    O. tokens / game
    107.5k
    glm-5.3-flash
  17. 17MiniMax M3 (Thinking)1261Elo
    Time / turn
    33.7s
    Cost / game
    $0.541
    Cheating %
    3.72%
    More statistics
    Wins %
    13%
    Draws %
    27%
    Losses %
    60%
    O. tokens / game
    165.9k
    minimax-m3
  18. 18Qwen3.7 Plus (Thinking)1075Elo
    Time / turn
    8.3s
    Cost / game
    $0.072
    Cheating %
    1.30%
    More statistics
    Wins %
    4%
    Draws %
    33%
    Losses %
    63%
    O. tokens / game
    13.5k
    qwen3.7-plus

Why is ChessBench a good benchmark?

Read the methodology

Reason beyond familiar tasks

Chess complements coding benchmarks by asking models to apply learned knowledge in a different setting. Each changing position tests how well they combine rules, constraints, and long-term goals.

Rules, constraints, and a goal inform a choice in the current position. The small grid represents a board detail.RulesConstraintsGoalCurrent positionChoose a moveRules, constraints, and a goal inform a choice in the current position. The small grid represents a board detail.RulesConstraintsGoalCurrentpositionChoose a move

Revise the plan. Keep the goal.

New context can expose gaps or incorrect assumptions in a coding plan. An unexpected chess reply can do the same to a planned sequence. In both cases, the model must reassess the situation, revise its approach, and keep working toward its goal.

Possible replies lead to an observed reply. The earlier plan ends, while new evidence leads to a revised plan.RulesConstraintsGoalCurrentpositionReply AReply BReply CObserved replyEarlier planUse new evidenceRevised planPossible replies lead to an observed reply. The earlier plan ends, while new evidence leads to a revised plan.RulesConstraintsGoalCurrentpositionReply AReply BReply CObserved replyEarlier planNew evidenceRevised plan

Think beyond the next move

A strong move accounts for several possible replies. The model must compare future scenarios, identify threats, and weigh consequences before acting. This kind of forward planning also matters in strategy, risk analysis, and forecasting.

A move branches into several possible replies and later consequences. A blue route marks one possible continuation.Your moveReply AReply BReply CCompare possible continuationsA move branches into several possible replies and later consequences. A blue route marks one possible continuation.Your moveReply AReply BReply CCompare possible continuations

Chess knowledge may be present in training data. Transfer to coding or stock-market forecasting needs separate evidence.