GPT-5.5 vs GPT-5.4
17 shared benchmarks
16
wins
1
ties
0
wins
85.0%
ARC-AGI 2
73.3%
0.43%
ARC-AGI 3
0.26%
46.9%
Agents' Last Exam
44.7%
84.4%
BrowseComp
82.7%
81.8%
CyberGym
79.0%
67.0%
DeepSWE
52.0%
51.7%
FrontierMath
47.6%
35.4%
FrontierMath Tier 4
27.1%
84.9%
GDPval
83.0%
93.5%
GPQA Diamond
92.8%
41.4%
Humanity's Last Exam
36.24%
81.2%
MMMU Pro
81.2%
74.0%
MRCR v2
36.6%
78.7%
OSWorld
75.0%
78.7%
OSWorld-Verified
75.0%
58.6%
SWE-bench Pro
57.7%
82.7%
Terminal-Bench 2.0
75.1%