Tihu · Pelican Bench

Models: 27 · Tiers: 54 · Samples: 165 · Invalid: 3

Day highlights

Gemini 4 Argon

8.83 · high (best)
Model page

high

8.83 ± 0.18

medium

8.77 ± 0.02

low

8.39 ± 0.29

GPT-6 Astra

8.79 · max (best)
Model page

max

8.79 ± 0.38

low

8.24 ± 0.06

medium

8.22 ± 0.78

Gemini 3.8 Flash

8.69 · high (best)
Model page

high

8.69 ± 0.35

medium

8.43 ± 0.41
Judge scores vary between passes

low

8.24 ± 0.32

GLM 5.3 Flash

8.57 · default (best)
Model page

default

8.57 ± 0.39

GPT-6.1 Sol

8.53 · max (best)
Model page

max

8.53 ± 0.34

low

8.23 ± 0.30

medium

7.28 ± 1.04

GLM 5.3

8.41 · default (best)
Model page

default

8.41 ± 0.42

GPT-6 Sol

8.40 · max (best)
Model page

max

8.40 ± 0.53

medium

6.26 ± 0.07

low

5.53 ± 0.06

Doubao Seed 2.1 Pro

8.21 · default (best)
Model page

default

8.21 ± 0.28

DeepSeek V4.1 Flash

8.20 · default (best)
Model page

default

8.20 ± 0.33

Qwen3.8-Max

8.16 · default (best)
Model page

default

8.16 ± 0.36

Qwen3.8-Flash

8.14 · default (best)
Model page

default

8.14 ± 0.77
Judge scores vary between passes

MiniMax-M3

7.99 · default (best)
Model page

default

7.99 ± 0.55

Claude Opus 5.5

7.94 · high (best)
Model page

high

7.94 ± 0.64
Judge scores vary between passes

medium

7.78 ± 0.22

low

6.10 ± 0.79

GPT-5.6 Sol

7.83 · max (best)
Model page

max

7.83 ± 1.09

medium

6.32 ± 0.89

low

6.15 ± 0.80

Kimi K3

7.73 · max (best)
Model page

max

7.73 ± 0.31

high

6.83 ± 0.44

low

5.26 ± 0.61

Gemini 3.1 Pro

7.44 · default (best)
Model page

default

7.44 ± 0.29

DeepSeek V4 Pro

7.27 · default (best)
Model page

default

7.27 ± 0.32
Judge scores vary between passes

GPT-5.6 Terra

7.23 · max (best)
Model page

max

7.23 ± 0.57

medium

6.47 ± 0.74

low

5.63 ± 1.16

Grok 4.7

6.99 · xhigh (best)
Model page

xhigh

6.99 ± 0.86

medium

6.02 ± 0.83

low

4.71 ± 1.20

GPT-6 Luna

6.90 · max (best)
Model page

max

6.90 ± 0.65

medium

5.36 ± 0.82

low

4.28 ± 0.33

Claude Sonnet 5.5

6.63 · high (best)
Model page

high

6.63 ± 0.38

low

5.70 ± 0.60

medium

5.60 ± 0.81

Grok 4.6

6.46 · high (best)
Model page

high

6.46 ± 0.95

GLM 5.2

6.06 · max (best)
Model page

max

6.06 ± 0.37
Judge scores vary between passes

high

5.03 ± 0.95

Hy3

5.95 · default (best)
Model page

default

5.95 ± 0.84

GPT-5.6 Luna

5.64 · max (best)
Model page

max

5.64 ± 0.14
Judge scores vary between passes

medium

4.60 ± 0.72

low

4.39 ± 0.42

Grok 4.5

5.59 · high (best)
Model page

high

5.59 ± 0.87

Composer 2.5

5.01 · default (best)
Model page

default

5.01 ± 0.60