Tihu · Pelican Bench

Models: 27 · Tiers: 54 · Samples: 170 · Invalid: 8

Day highlights

Gemini 4 Argon

8.86 · high (best)
Model page

high

8.86 ± 0.11

medium

8.76 ± 0.11

low

8.45 ± 0.26

GPT-6 Astra

8.79 · max (best)
Model page

max

8.79 ± 0.22

low

8.64 ± 0.06

medium

8.53 ± 0.27

GPT-6.1 Sol

8.75 · max (best)
Model page

max

8.75 ± 0.14

medium

7.86 ± 0.69

low

7.52 ± 0.12

Qwen3.8-Max

8.74 · default (best)
Model page

default

8.74 ± 0.14

Qwen3.8-Flash

8.64 · default (best)
Model page

default

8.64 ± 0.19

GLM 5.3

8.50 · default (best)
Model page

default

8.50 ± 0.29

GPT-6 Sol

8.32 · max (best)
Model page

max

8.32 ± 0.68
Judge scores vary between passes

low

6.57 ± 0.21
Judge scores vary between passes

medium

5.98 ± 0.81

Gemini 3.8 Flash

8.21 · high (best)
Model page

high

8.21 ± 0.78

medium

7.56 ± 0.69

low

7.46 ± 0.86
Judge scores vary between passes

Kimi K3

7.89 · max (best)
Model page

max

7.89 ± 0.71

high

6.87 ± 0.32

low

5.46 ± 0.58

Claude Opus 5.5

7.77 · medium (best)
Model page

medium

7.77 ± 0.79
Judge scores vary between passes

high

7.48 ± 0.70
Judge scores vary between passes

low

6.06 ± 0.73

GLM 5.3 Flash

7.60 · default (best)
Model page

default

7.60 ± 0.45

Gemini 3.1 Pro

7.40 · default (best)
Model page

default

7.40 ± 0.92

GPT-6 Luna

7.38 · max (best)
Model page

max

7.38 ± 0.76

medium

6.54 ± 1.55

low

5.95 ± 0.25
Judge scores vary between passes
Judge scores vary between passes

Doubao Seed 2.1 Pro

7.04 · default (best)
Model page

default

7.04 ± 0.48

Hy3

6.94 · default (best)
Model page

default

6.94 ± 1.16
Judge scores vary between passes

DeepSeek V4.1 Flash

6.88 · default (best)
Model page

default

6.88 ± 0.29

GPT-5.6 Sol

6.77 · low (best)
Model page

low

6.77 ± 1.06

max

6.48 ± 1.87

medium

6.28 ± 1.58

DeepSeek V4 Pro

6.76 · default (best)
Model page

default

6.76 ± 0.95

Claude Sonnet 5.5

6.70 · high (best)
Model page

high

6.70 ± 1.26

low

5.41 ± 0.43

medium

5.36 ± 0.91

GPT-5.6 Terra

6.62 · max (best)
Model page

max

6.62 ± 1.16
Judge scores vary between passes

low

6.13 ± 0.43

medium

4.99 ± 0.21

Grok 4.7

6.40 · medium (best)
Model page

medium

6.40 ± 1.64
Judge scores vary between passes

xhigh

6.29 ± 1.41

low

5.70 ± 1.79

GLM 5.2

6.32 · max (best)
Model page

max

6.32 ± 0.58

high

4.86 ± 0.35

GPT-5.6 Luna

6.13 · max (best)
Model page

max

6.13 ± 1.88

medium

5.60 ± 0.88

low

4.32 ± 0.41

Grok 4.6

5.95 · high (best)
Model page

high

5.95 ± 0.68

MiniMax-M3

5.86 · default (best)
Model page

default

5.86 ± 1.62
Judge scores vary between passes

Grok 4.5

5.69 · high (best)
Model page

high

5.69 ± 0.71

Composer 2.5

4.76 · default (best)
Model page

default

4.76 ± 1.34