Compare · ModelsLive · 2 picked · head to head
GPT-5.1 vs GPT-5.2
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
GPT-5.2 wins on 19/24 benchmarks
GPT-5.2 wins 19 of 24 shared benchmarks. Leads in agentic · reasoning · arena.
Category leads
agentic·GPT-5.2reasoning·GPT-5.2arena·GPT-5.2knowledge·GPT-5.2general·GPT-5.2math·GPT-5.2coding·GPT-5.2
Hype vs Reality
Attention vs performance
GPT-5.1
#157 by perf·#9 by attention
GPT-5.2
#168 by perf·#9 by attention
Best value
GPT-5.1
1.5x better value than GPT-5.2
GPT-5.1
8.4 pts/$
$5.63/M
GPT-5.2
5.8 pts/$
$7.88/M
Vendor risk
Who is behind the model
OpenAI
$840.0B·Tier 1
OpenAI
$840.0B·Tier 1
Head to head
24 benchmarks · 2 models
GPT-5.1GPT-5.2
APEX-Agents
GPT-5.2 leads by +16.8
APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments.
GPT-5.1
17.5
GPT-5.2
34.3
ARC-AGI
GPT-5.2 leads by +13.4
ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.
GPT-5.1
72.8
GPT-5.2
86.2
ARC-AGI-2
GPT-5.2 leads by +35.3
ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data.
GPT-5.1
17.6
GPT-5.2
52.9
Chatbot Arena Elo · Coding
GPT-5.2 leads by +74.4
GPT-5.1
1341.0
GPT-5.2
1415.5
Chatbot Arena Elo · Overall
GPT-5.1 leads by +2.9
GPT-5.1
1438.5
GPT-5.2
1435.6
Chess Puzzles
GPT-5.2 leads by +6.3
Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities.
GPT-5.1
39.0
GPT-5.2
45.3
Cl Bench
GPT-5.1 leads by +5.5
GPT-5.1
23.7
GPT-5.2
18.2
DeepResearch Bench
GPT-5.1 leads by +1.7
DeepResearch Bench · evaluates AI on complex multi-step research tasks requiring information gathering, synthesis, and producing comprehensive analyses.
GPT-5.1
42.8
GPT-5.2
41.1
Dtbench
GPT-5.2 leads by +1.3
GPT-5.1
83.5
GPT-5.2
84.9
FrontierMath-2025-02-28-Private
GPT-5.2 leads by +9.7
FrontierMath (Feb 2025) · original research-level math problems created by mathematicians, testing capabilities at the boundary of current AI mathematical reasoning.
GPT-5.1
31.0
GPT-5.2
40.7
FrontierMath-Tier-4-2025-07-01-Private
GPT-5.2 leads by +6.3
FrontierMath Tier 4 (Jul 2025) · the most challenging tier of frontier mathematics, containing problems that push the absolute limits of AI mathematical reasoning.
GPT-5.1
12.5
GPT-5.2
18.8
GPQA diamond
GPT-5.2 leads by +5.2
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
GPT-5.1
87.0
GPT-5.2
92.2
GSO-Bench
GPT-5.2 leads by +13.7
GSO-Bench · evaluates AI models on real-world open-source software engineering tasks, testing the ability to understand and resolve actual GitHub issues.
GPT-5.1
13.7
GPT-5.2
27.4
HLE
GPT-5.2 leads by +4.3
HLE (Humanity's Last Exam) · a reasoning benchmark designed to be the hardest public evaluation of AI. Questions span mathematics, physics, philosophy, and logic · curated to be at or beyond the frontier of human expert capability. Tested with and without tool augmentation.
GPT-5.1
19.8
GPT-5.2
24.2
Lmca
GPT-5.2 leads by +0.0
GPT-5.1
51.6
GPT-5.2
51.7
Mystery Game Puzzles
GPT-5.2 leads by +4.4
GPT-5.1
10.8
GPT-5.2
15.2
OTIS Mock AIME 2024-2025
GPT-5.2 leads by +7.5
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
GPT-5.1
88.6
GPT-5.2
96.1
SimpleBench
GPT-5.1 leads by +8.9
SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking.
GPT-5.1
43.8
GPT-5.2
35.0
SimpleQA Verified
GPT-5.1 leads by +9.1
SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information.
GPT-5.1
48.0
GPT-5.2
38.9
SWE-Bench verified
GPT-5.2 leads by +5.8
SWE-bench Verified · 500 human-validated tasks from 12 real Python repositories (Django, Flask, scikit-learn, sympy, and others). Each task requires the model to produce a git patch that resolves a real GitHub issue and passes the test suite. The verified subset eliminates ambiguous tasks from the original SWE-bench. The benchmark remains the most-cited evaluation for code-generation capability.
GPT-5.1
68.0
GPT-5.2
73.8
SWE-Bench Verified (Bash Only)
GPT-5.2 leads by +5.8
SWE-Bench Verified (Bash Only) · a curated subset of SWE-bench where models fix real Python repository bugs using only bash commands, no agent frameworks.
GPT-5.1
66.0
GPT-5.2
71.8
Terminal Bench
GPT-5.2 leads by +17.3
Terminal-Bench 2.0 · evaluates AI agents on real terminal-based coding tasks · writing scripts, debugging, running tests, and managing projects entirely through command-line interaction. Tests both code quality and terminal fluency.
GPT-5.1
47.6
GPT-5.2
64.9
VPCT
GPT-5.2 leads by +38.0
VPCT (Visual Pattern Completion Test) · tests visual reasoning and pattern recognition by having models complete visual sequences and transformations.
GPT-5.1
38.0
GPT-5.2
76.0
WeirdML
GPT-5.2 leads by +11.4
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
GPT-5.1
60.8
GPT-5.2
72.2
Full benchmark table
| Benchmark | GPT-5.1 | GPT-5.2 |
|---|---|---|
APEX-Agents APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments. | 17.5 | 34.3 |
ARC-AGI ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization. | 72.8 | 86.2 |
ARC-AGI-2 ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data. | 17.6 | 52.9 |
Chatbot Arena Elo · Coding | 1341.0 | 1415.5 |
Chatbot Arena Elo · Overall | 1438.5 | 1435.6 |
Chess Puzzles Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities. | 39.0 | 45.3 |
Cl Bench | 23.7 | 18.2 |
DeepResearch Bench DeepResearch Bench · evaluates AI on complex multi-step research tasks requiring information gathering, synthesis, and producing comprehensive analyses. | 42.8 | 41.1 |
Dtbench | 83.5 | 84.9 |
FrontierMath-2025-02-28-Private FrontierMath (Feb 2025) · original research-level math problems created by mathematicians, testing capabilities at the boundary of current AI mathematical reasoning. | 31.0 | 40.7 |
FrontierMath-Tier-4-2025-07-01-Private FrontierMath Tier 4 (Jul 2025) · the most challenging tier of frontier mathematics, containing problems that push the absolute limits of AI mathematical reasoning. | 12.5 | 18.8 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 87.0 | 92.2 |
GSO-Bench GSO-Bench · evaluates AI models on real-world open-source software engineering tasks, testing the ability to understand and resolve actual GitHub issues. | 13.7 | 27.4 |
HLE HLE (Humanity's Last Exam) · a reasoning benchmark designed to be the hardest public evaluation of AI. Questions span mathematics, physics, philosophy, and logic · curated to be at or beyond the frontier of human expert capability. Tested with and without tool augmentation. | 19.8 | 24.2 |
Lmca | 51.6 | 51.7 |
Mystery Game Puzzles | 10.8 | 15.2 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 88.6 | 96.1 |
SimpleBench SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking. | 43.8 | 35.0 |
SimpleQA Verified SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information. | 48.0 | 38.9 |
SWE-Bench verified SWE-bench Verified · 500 human-validated tasks from 12 real Python repositories (Django, Flask, scikit-learn, sympy, and others). Each task requires the model to produce a git patch that resolves a real GitHub issue and passes the test suite. The verified subset eliminates ambiguous tasks from the original SWE-bench. The benchmark remains the most-cited evaluation for code-generation capability. | 68.0 | 73.8 |
SWE-Bench Verified (Bash Only) SWE-Bench Verified (Bash Only) · a curated subset of SWE-bench where models fix real Python repository bugs using only bash commands, no agent frameworks. | 66.0 | 71.8 |
Terminal Bench Terminal-Bench 2.0 · evaluates AI agents on real terminal-based coding tasks · writing scripts, debugging, running tests, and managing projects entirely through command-line interaction. Tests both code quality and terminal fluency. | 47.6 | 64.9 |
VPCT VPCT (Visual Pattern Completion Test) · tests visual reasoning and pattern recognition by having models complete visual sequences and transformations. | 38.0 | 76.0 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 60.8 | 72.2 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
People also compared