Tihu · Pelican Bench

Models: 10 · Tiers: 26 · Samples: 79 · Invalid: 1

Day highlights

GPT-6 Astra

8.82 · max (best)
Model page

max

8.82 ± 0.10

medium

8.40 ± 0.25

low

7.04 ± 0.59
Judge scores vary between passes

GPT-6.1 Sol

8.58 · max (best)
Model page

max

8.58 ± 0.20

medium

8.20 ± 0.13

low

7.16 ± 0.21

GPT-5.6 Sol

8.17 · max (best)
Model page

max

8.17 ± 0.38

medium

6.68 ± 1.65

low

6.56 ± 0.41

GPT-6 Luna

7.48 · max (best)
Model page

max

7.48 ± 0.42

low

6.50 ± 0.96
Judge scores vary between passes

medium

5.71 ± 0.38
Judge scores vary between passes

GPT-5.6 Terra

7.46 · max (best)
Model page

max

7.46 ± 0.69

low

7.44 ± 0.69
Judge scores vary between passes

medium

6.22 ± 1.69

GPT-6 Sol

7.40 · max (best)
Model page

max

7.40 ± 0.90

medium

6.34 ± 0.28
Judge scores vary between passes

low

6.07 ± 1.12

Grok 4.7

7.00 · xhigh (best)
Model page

xhigh

7.00 ± 0.79

low

5.59 ± 0.85

medium

5.49 ± 0.13

Grok 4.6

6.97 · high (best)
Model page

high

6.97 ± 0.13

GPT-5.6 Luna

6.32 · max (best)
Model page

max

6.32 ± 1.06

low

5.24 ± 0.17

medium

5.00 ± 0.93

Grok 4.5

6.30 · high (best)
Model page

high

6.30 ± 0.85