Tihu · Pelican Bench

Models: 10 · Tiers: 26 · Samples: 85 · Invalid: 7

Day highlights

GPT-6.1 Sol

8.62 · max (best)
Model page

max

8.62 ± 0.39

low

8.11 ± 0.23

medium

7.03 ± 0.37
Judge scores vary between passes

GPT-6 Astra

8.55 · max (best)
Model page

max

8.55 ± 0.63

medium

7.48 ± 1.29

low

7.04 ± 0.92

Grok 4.6

7.89 · high (best)
Model page

high

7.89 ± 0.18

GPT-5.6 Sol

7.80 · max (best)
Model page

max

7.80 ± 0.43

low

7.15 ± 0.61
Judge scores vary between passes

medium

6.78 ± 1.54

GPT-5.6 Terra

7.39 · max (best)
Model page

max

7.39 ± 0.89

low

6.09 ± 1.39

medium

6.08 ± 0.78
Judge scores vary between passes

GPT-5.6 Luna

7.27 · max (best)
Model page

max

7.27 ± 0.77

medium

5.71 ± 1.43

low

4.47 ± 1.03

GPT-6 Sol

6.97 · max (best)
Model page

max

6.97 ± 0.69
Judge scores vary between passes

low

6.73 ± 1.03

medium

6.55 ± 0.27

GPT-6 Luna

6.70 · max (best)
Model page

max

6.70 ± 0.90

medium

6.24 ± 1.04

low

5.62 ± 1.16

Grok 4.5

6.08 · high (best)
Model page

high

6.08 ± 0.31

Grok 4.7

5.58 · medium (best)
Model page

medium

5.58 ± 1.23

xhigh

5.30 ± 1.34

low

5.21 ± 0.16