20251212-103449
1
Slide 1
NotebookLM to PPTXTurn NotebookLM Slides into Editable PPTXExport PPTX
Log In
59.6%
70.9%
70.9%
31.1%
17.6%
37.6%
52.9%
52.9%
75.0%
75.0%
72.8%
75.0%
80.6%
72.8%
88.2%
86.2%
95.0%
94.0%
95.0%
100.0%
96.0%
92.6%
92.8%
94.0%
94.0%
100.0%
37.8%
31.0%
40.3%
81.4%
67.0%
82.1%
67.0%
87.0%
81.4%
81.4%
82.1%
88.1%
92.4%
91.9%
88.1%
43.3%
57.0%
92.4%
50.8%
91.9%
52.0%
92.4%
55.6%
43.3%
50.8%
55.6%
shows dominant
(Knowledge Work)
GDPval
(Abstract Reasoning)
ARC-AGI2
(Abstract Reasoning)
ARC-AGI1
(Competition Math)
AIME 2025
=t16%-
a2+4a
m=im
(Advanced Math)
FrontierMath
CharXiv Reasoning (Scientific Figures)
(Science Questions)
GPQA Diamond
(Software Engineering)
SWE-Bench Pro
Massive Leap in Abstract Reasoning
KEY FINDINGS
This infographic compares benchmark scores of leading Al models, including OpenAl's GPT-5.2, Anthropic's Claude Opus 4.5, and Google's Gemini 3 Pro, on complex tasks in software engineering, math, science, and abstract problem-solving.
A New Leader in Al Reasoning
GPT-5.2:
NotebookLM
38.8%
53.5%
59.6%
GPT-5.2 scored significantly higher than all other models on ARC-AGI benchmarks.
It achieves the highest score in every single benchmark category presented.
performance across the board.
GPT-5.2