Tihu · Pelican Bench

Data date:

Models: 30 · Tiers: 59 · Samples: 179 · Invalid: 2

Day highlights

Gemini 4 Argon

8.97 · medium (best)
Model page

medium

8.97 ± 0.04

high

8.85 ± 0.09

low

8.46 ± 0.59

GPT-6 Astra

8.68 · max (best)
Model page

max

8.68 ± 0.12

medium

8.41 ± 0.16

low

7.43 ± 0.78

GPT-6.1 Sol

8.66 · max (best)
Model page

max

8.66 ± 0.27

medium

8.09 ± 1.01

low

7.94 ± 0.69

Qwen3.8-Flash

8.43 · default (best)
Model page

default

8.43 ± 0.49

GLM 5.3

8.36 · default (best)
Model page

default

8.36 ± 0.28

Claude Opus 5.5

8.32 · high (best)
Model page

high

8.32 ± 0.30

low

7.02 ± 0.34

medium

6.52 ± 0.91

Kimi K3

8.26 · max (best)
Model page

max

8.26 ± 0.30

high

6.99 ± 0.57

low

6.19 ± 0.58

MiniMax M3.1 Flash

8.21 · default (best)
Model page

default

8.21 ± 0.49

Gemini 3.8 Flash

8.20 · high (best)
Model page

high

8.20 ± 0.44
Judge scores vary between passes

low

8.11 ± 0.85

medium

7.47 ± 0.96
Judge scores vary between passes

Qwen3.8-Max

8.16 · default (best)
Model page

default

8.16 ± 0.23

MiniMax-M3

7.99 · default (best)
Model page

default

7.99 ± 0.38

GLM 5.3 Flash

7.93 · default (best)
Model page

default

7.93 ± 0.16
Judge scores vary between passes

GPT-5.6 Sol

7.55 · max (best)
Model page

max

7.55 ± 1.13
Judge scores vary between passes

medium

7.35 ± 0.51

low

6.10 ± 0.46

Kimi K3.1 Pro

7.40 · default (best)
Model page

default

7.40 ± 0.89
Judge scores vary between passes

GPT-6.1 Terra

7.27 · max (best)
Model page

max

7.27 ± 1.01

low

6.95 ± 0.47

medium

6.04 ± 0.83
Judge scores vary between passes

DeepSeek V4.1 Flash

7.21 · default (best)
Model page

default

7.21 ± 0.88

Grok 4.6

7.06 · high (best)
Model page

high

7.06 ± 0.40

GPT-6 Luna

6.98 · max (best)
Model page

max

6.98 ± 1.32

medium

6.64 ± 1.02

low

6.22 ± 0.63
Judge scores vary between passes

Hy3

6.85 · default (best)
Model page

default

6.85 ± 0.85

Grok 4.7

6.81 · xhigh (best)
Model page

xhigh

6.81 ± 0.70

medium

5.49 ± 0.23

low

5.41 ± 0.37

DeepSeek V4 Pro

6.73 · default (best)
Model page

default

6.73 ± 0.95
Judge scores vary between passes

Gemini 3.1 Pro

6.73 · default (best)
Model page

default

6.73 ± 1.23

GPT-6 Sol

6.68 · max (best)
Model page

max

6.68 ± 0.11
Judge scores vary between passes

medium

6.65 ± 0.71

low

5.78 ± 0.86

Claude Sonnet 5.5

6.60 · high (best)
Model page

high

6.60 ± 0.85

low

6.09 ± 1.10

medium

5.67 ± 0.47

Doubao Seed 2.1 Pro

6.27 · default (best)
Model page

default

6.27 ± 0.68

Grok 4.5

6.19 · high (best)
Model page

high

6.19 ± 0.96

GPT-5.6 Terra

6.13 · max (best)
Model page

max

6.13 ± 0.77

low

5.57 ± 0.35

medium

5.50 ± 1.06
Judge scores vary between passes

GLM 5.2

6.04 · max (best)
Model page

max

6.04 ± 0.21

high

5.51 ± 0.79

GPT-5.6 Luna

5.95 · max (best)
Model page

max

5.95 ± 1.13

medium

5.32 ± 0.52
Judge scores vary between passes

low

4.08 ± 0.72

Composer 2.5

4.81 · default (best)
Model page

default

4.81 ± 0.64