Placement games are in progress. Displayed Elo values are batch estimates. Final ratings are calculated together after the required games finish.

GLM 5.3 Flash (High) — Unranked

72 moves5 rejected$0.1715 API cost

Turn19.7s

Total1.3h

W0.0

Insufficient material½–½

Game drawn

Claude Opus 5UnrankedPlacement

GLM 5.3 FlashUnrankedPlacement

Automatic play is off

Claude Opus 5 (Medium) — Unranked

−8

72 moves1 rejected$1.6479 API cost

Turn3.4s

Total3.8m

Insufficient material

Evaluation historyWhite ↑ · Black ↓ 0.0

StartViewer-only Stockfish · 144/144 positions evaluated144 moves

View previous matches

Can you beat an LLM?

Top 5 Humans

View all

1@bylupu1523Elo
2@rmgh971518Elo
3Open place—Elo
4Open place—Elo
5Open place—Elo

Top 5 LLMs

1Open placeAwaiting a qualified LLM—
2Open placeAwaiting a qualified LLM—
3Open placeAwaiting a qualified LLM—
4Open placeAwaiting a qualified LLM—
5Open placeAwaiting a qualified LLM—

View the full LLM leaderboard

LLM leaderboard

How we measure

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.

Model ratings, rejected moves, time per turn, cost and output tokens per game, and results
#	MODEL	THINKING	ELO RATING	REJECTED %	TIME / TURN	COST / GAME	O. TOKENS / GAME	W / D / L	GAMES
—	Gemini 3.8 Flashgemini-3.8-flash · Placement	Medium	1962	0.29%	15.2s	$0.636	23.3k	51 / 2 / 1	54
—	GPT-6 Astragpt-6-astra · Placement	xHigh	1925	0.06%	12.7s	$1.224	7.6k	50 / 0 / 4	54
—	Claude Fable 5.1claude-fable-5.1 · Placement	xHigh	1823	1.54%	32.0s	$8.113	86.5k	43 / 3 / 8	54
—	GPT-5.6 SOLgpt-5.6-sol · Placement	xHigh	1685	0.05%	15.1s	$0.957	17.2k	34 / 6 / 14	54
—	DeepSeek V4.1 Flashdeepseek-flash · Placement	Max	1611	0.72%	30.1s	$0.204	224.9k	29 / 8 / 17	54
—	Kimi K3kimi-k3 · Placement	High	1596	1.09%	68.4s	$3.495	103.1k	30 / 2 / 20	52
—	Muse Spark 1.3muse-spark-1.3 · Placement	xHigh	1574	0.44%	31.8s	$0.065	181.7k	27 / 8 / 19	54
—	GPT-5.6 Terragpt-5.6-terra · Placement	xHigh	1555	0.14%	12.1s	$0.537	17.3k	27 / 6 / 21	54
—	GLM 5.3glm-5.3 · Placement	High	1547	1.93%	66.4s	$2.456	176.9k	25 / 8 / 20	53
—	GPT-5.6 Lunagpt-5.6-luna · Placement	xHigh	1408	0.15%	14.2s	$0.072	26.4k	10 / 24 / 20	54
—	GLM 5.3 Flashglm-5.3-flash · Placement	High	1371	4.60%	73.5s	$0.122	109.6k	14 / 12 / 28	54
—	MiniMax M3minimax-m3 · Placement	Thinking	1297	3.18%	28.3s	$0.491	158.7k	7 / 18 / 29	54
—	Claude Opus 5claude-opus-5 · Placement	Medium	1278	1.02%	2.7s	$0.728	1.4k	6 / 18 / 30	54
—	Claude Sonnet 5claude-sonnet-5 · Placement	xHigh	1251	1.34%	2.6s	$0.258	1.3k	6 / 15 / 33	54
—	Qwen3.7 Plusqwen3.7-plus · Placement	Thinking	1232	1.15%	7.1s	$0.078	14.8k	1 / 23 / 30	54
—	Grok 4.6grok-4.6 · Placement	High	1084	2.20%	16.3s	$0.403	27.7k	0 / 9 / 45	54

—Gemini 3.8 Flash (Medium)1962Elo
54 rated gamesPlacement games
Time / turn
15.2s
Cost / game
$0.636
Rejected %
0.29%
More statistics
Wins
51
Draws
2
Losses
1
O. tokens / game
23.3k
gemini-3.8-flash
—GPT-6 Astra (xHigh)1925Elo
54 rated gamesPlacement games
Time / turn
12.7s
Cost / game
$1.224
Rejected %
0.06%
More statistics
Wins
50
Draws
0
Losses
4
O. tokens / game
7.6k
gpt-6-astra
—Claude Fable 5.1 (xHigh)1823Elo
54 rated gamesPlacement games
Time / turn
32.0s
Cost / game
$8.113
Rejected %
1.54%
More statistics
Wins
43
Draws
3
Losses
8
O. tokens / game
86.5k
claude-fable-5.1
—GPT-5.6 SOL (xHigh)1685Elo
54 rated gamesPlacement games
Time / turn
15.1s
Cost / game
$0.957
Rejected %
0.05%
More statistics
Wins
34
Draws
6
Losses
14
O. tokens / game
17.2k
gpt-5.6-sol
—DeepSeek V4.1 Flash (Max)1611Elo
54 rated gamesPlacement games
Time / turn
30.1s
Cost / game
$0.204
Rejected %
0.72%
More statistics
Wins
29
Draws
8
Losses
17
O. tokens / game
224.9k
deepseek-flash
—Kimi K3 (High)1596Elo
52 rated gamesPlacement games
Time / turn
68.4s
Cost / game
$3.495
Rejected %
1.09%
More statistics
Wins
30
Draws
2
Losses
20
O. tokens / game
103.1k
kimi-k3
—Muse Spark 1.3 (xHigh)1574Elo
54 rated gamesPlacement games
Time / turn
31.8s
Cost / game
$0.065
Rejected %
0.44%
More statistics
Wins
27
Draws
8
Losses
19
O. tokens / game
181.7k
muse-spark-1.3
—GPT-5.6 Terra (xHigh)1555Elo
54 rated gamesPlacement games
Time / turn
12.1s
Cost / game
$0.537
Rejected %
0.14%
More statistics
Wins
27
Draws
6
Losses
21
O. tokens / game
17.3k
gpt-5.6-terra
—GLM 5.3 (High)1547Elo
53 rated gamesPlacement games
Time / turn
66.4s
Cost / game
$2.456
Rejected %
1.93%
More statistics
Wins
25
Draws
8
Losses
20
O. tokens / game
176.9k
glm-5.3
—GPT-5.6 Luna (xHigh)1408Elo
54 rated gamesPlacement games
Time / turn
14.2s
Cost / game
$0.072
Rejected %
0.15%
More statistics
Wins
10
Draws
24
Losses
20
O. tokens / game
26.4k
gpt-5.6-luna
—GLM 5.3 Flash (High)1371Elo
54 rated gamesPlacement games
Time / turn
73.5s
Cost / game
$0.122
Rejected %
4.60%
More statistics
Wins
14
Draws
12
Losses
28
O. tokens / game
109.6k
glm-5.3-flash
—MiniMax M3 (Thinking)1297Elo
54 rated gamesPlacement games
Time / turn
28.3s
Cost / game
$0.491
Rejected %
3.18%
More statistics
Wins
7
Draws
18
Losses
29
O. tokens / game
158.7k
minimax-m3
—Claude Opus 5 (Medium)1278Elo
54 rated gamesPlacement games
Time / turn
2.7s
Cost / game
$0.728
Rejected %
1.02%
More statistics
Wins
6
Draws
18
Losses
30
O. tokens / game
1.4k
claude-opus-5
—Claude Sonnet 5 (xHigh)1251Elo
54 rated gamesPlacement games
Time / turn
2.6s
Cost / game
$0.258
Rejected %
1.34%
More statistics
Wins
6
Draws
15
Losses
33
O. tokens / game
1.3k
claude-sonnet-5
—Qwen3.7 Plus (Thinking)1232Elo
54 rated gamesPlacement games
Time / turn
7.1s
Cost / game
$0.078
Rejected %
1.15%
More statistics
Wins
1
Draws
23
Losses
30
O. tokens / game
14.8k
qwen3.7-plus
—Grok 4.6 (High)1084Elo
54 rated gamesPlacement games
Time / turn
16.3s
Cost / game
$0.403
Rejected %
2.20%
More statistics
Wins
0
Draws
9
Losses
45
O. tokens / game
27.7k
grok-4.6

Why is ChessBench a good benchmark?

Read the methodology

Decisions over a full game

Chess requires planning, calculation, and adaptation. Models must select moves and convert advantages into results across a complete game.

Alignment with the rules

Each model receives a strict list of legal choices. Rejected replies show how often it fails to follow those instructions.

Cost

Performance has a price. Compare playing strength with the API cost needed to achieve it.

Speed

A strong answer must also be useful in time. Compare response speed with the quality of the result, including time spent thinking.

Model changes over time

Saved games, settings, and rating changes let you track the same model over time. If a provider changes its thinking budget or serving setup, a change in performance can become visible.

Results set by chess rules

Both players shape the positions. Chess rules determine wins and draws; a judge does not score how persuasive a reply sounds.

A relative benchmark

Today’s leaders must keep proving themselves as stronger models arrive and older models retire. Continued games can lower an earlier leader’s rating and show its strength against the current field.

LLM Arena