Tihu · Pelican Bench

Models: 10 · Tiers: 30 · Samples: 90 · Invalid: 0

Day highlights

GPT-6 Astra

8.70 · max (best)
Model page

max

8.70 ± 0.27

medium

7.51 ± 0.51
Judge scores vary between passes

low

7.22 ± 1.00

GPT-6.1 Sol

8.24 · max (best)
Model page

max

8.24 ± 0.23

low

7.80 ± 0.64

medium

7.69 ± 0.34
Judge scores vary between passes
Judge scores vary between passes

Claude Opus 5.5

7.92 · medium (best)
Model page

medium

7.92 ± 0.18

high

7.15 ± 0.34
Judge scores vary between passes

low

6.17 ± 0.63

GPT-5.6 Sol

7.55 · max (best)
Model page

max

7.55 ± 1.22

low

7.18 ± 1.43

medium

7.02 ± 0.56
Judge scores vary between passes
Judge scores vary between passes

GPT-6 Sol

7.30 · medium (best)
Model page

medium

7.30 ± 1.07

max

7.27 ± 0.69

low

4.95 ± 0.54

Claude Sonnet 5.5

7.28 · high (best)
Model page

high

7.28 ± 0.67

medium

6.65 ± 0.83

low

6.41 ± 0.75

GPT-6.1 Terra

7.26 · max (best)
Model page

max

7.26 ± 0.62

medium

6.40 ± 1.14

low

6.23 ± 1.31

GPT-5.6 Terra

6.81 · max (best)
Model page

max

6.81 ± 1.39

medium

6.63 ± 1.48

low

6.39 ± 1.27
Judge scores vary between passes

GPT-5.6 Luna

6.41 · max (best)
Model page

max

6.41 ± 1.45

low

5.55 ± 1.04

medium

5.46 ± 1.11

GPT-6 Luna

6.35 · medium (best)
Model page

medium

6.35 ± 0.94

low

5.57 ± 0.68

max

5.18 ± 0.52