Compare · ModelsLive · 2 picked · head to head
Claude Sonnet 4.5 vs GPT-5
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
GPT-5 wins on 15/18 benchmarks
GPT-5 wins 15 of 18 shared benchmarks. Leads in reasoning · knowledge · math.
Category leads
reasoning·GPT-5knowledge·GPT-5math·GPT-5coding·GPT-5
Hype vs Reality
Attention vs performance
Claude Sonnet 4.5
#130 by perf·no signal
GPT-5
#72 by perf·no signal
Best value
GPT-5
2.1x better value than Claude Sonnet 4.5
Claude Sonnet 4.5
4.7 pts/$
$9.00/M
GPT-5
9.7 pts/$
$5.63/M
Vendor risk
Who is behind the model
Anthropic
$380.0B·Tier 1
OpenAI
$840.0B·Tier 1
Head to head
18 benchmarks · 2 models
Claude Sonnet 4.5GPT-5
ARC-AGI
GPT-5 leads by +2.0
ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.
Claude Sonnet 4.5
63.7
GPT-5
65.7
ARC-AGI-2
Claude Sonnet 4.5 leads by +3.8
ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data.
Claude Sonnet 4.5
13.6
GPT-5
9.9
Chess Puzzles
GPT-5 leads by +25.0
Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities.
Claude Sonnet 4.5
12.0
GPT-5
37.0
DeepResearch Bench
GPT-5 leads by +2.5
DeepResearch Bench · evaluates AI on complex multi-step research tasks requiring information gathering, synthesis, and producing comprehensive analyses.
Claude Sonnet 4.5
52.6
GPT-5
55.1
FrontierMath-2025-02-28-Private
GPT-5 leads by +17.2
FrontierMath (Feb 2025) · original research-level math problems created by mathematicians, testing capabilities at the boundary of current AI mathematical reasoning.
Claude Sonnet 4.5
15.2
GPT-5
32.4
FrontierMath-Tier-4-2025-07-01-Private
GPT-5 leads by +8.3
FrontierMath Tier 4 (Jul 2025) · the most challenging tier of frontier mathematics, containing problems that push the absolute limits of AI mathematical reasoning.
Claude Sonnet 4.5
4.2
GPT-5
12.5
GPQA diamond
GPT-5 leads by +5.1
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
Claude Sonnet 4.5
76.4
GPT-5
81.6
GSO-Bench
Claude Sonnet 4.5 leads by +7.8
GSO-Bench · evaluates AI models on real-world open-source software engineering tasks, testing the ability to understand and resolve actual GitHub issues.
Claude Sonnet 4.5
14.7
GPT-5
6.9
HLE
GPT-5 leads by +12.2
HLE (Humanity's Last Exam) · crowdsourced expert-level questions designed to be among the hardest possible challenges for AI systems across all domains.
Claude Sonnet 4.5
9.4
GPT-5
21.6
MATH level 5
GPT-5 leads by +0.4
MATH Level 5 · the hardest tier of the MATH benchmark, featuring competition-level problems from AMC, AIME, and Olympiad-style mathematics.
Claude Sonnet 4.5
97.7
GPT-5
98.1
OTIS Mock AIME 2024-2025
GPT-5 leads by +13.6
OTIS Mock AIME 2024–2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Claude Sonnet 4.5
77.8
GPT-5
91.4
SimpleBench
GPT-5 leads by +2.9
SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking.
Claude Sonnet 4.5
45.2
GPT-5
48.0
SimpleQA Verified
GPT-5 leads by +27.0
SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information.
Claude Sonnet 4.5
23.6
GPT-5
50.6
SWE-Bench verified
GPT-5 leads by +2.3
Claude Sonnet 4.5
71.3
GPT-5
73.5
SWE-Bench Verified (Bash Only)
Claude Sonnet 4.5 leads by +5.6
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.
Claude Sonnet 4.5
70.6
GPT-5
65.0
Terminal Bench
GPT-5 leads by +3.1
Terminal Bench · tests the ability to accomplish real-world tasks using terminal commands, evaluating shell scripting and CLI tool proficiency.
Claude Sonnet 4.5
46.5
GPT-5
49.6
VPCT
GPT-5 leads by +39.3
VPCT (Visual Pattern Completion Test) · tests visual reasoning and pattern recognition by having models complete visual sequences and transformations.
Claude Sonnet 4.5
9.7
GPT-5
49.0
WeirdML
GPT-5 leads by +13.0
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
Claude Sonnet 4.5
47.7
GPT-5
60.7
Full benchmark table
| Benchmark | Claude Sonnet 4.5 | GPT-5 |
|---|---|---|
ARC-AGI ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization. | 63.7 | 65.7 |
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. | 13.6 | 9.9 |
Chess Puzzles Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities. | 12.0 | 37.0 |
DeepResearch Bench DeepResearch Bench · evaluates AI on complex multi-step research tasks requiring information gathering, synthesis, and producing comprehensive analyses. | 52.6 | 55.1 |
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. | 15.2 | 32.4 |
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. | 4.2 | 12.5 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 76.4 | 81.6 |
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. | 14.7 | 6.9 |
HLE HLE (Humanity's Last Exam) · crowdsourced expert-level questions designed to be among the hardest possible challenges for AI systems across all domains. | 9.4 | 21.6 |
MATH level 5 MATH Level 5 · the hardest tier of the MATH benchmark, featuring competition-level problems from AMC, AIME, and Olympiad-style mathematics. | 97.7 | 98.1 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024–2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 77.8 | 91.4 |
SimpleBench SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking. | 45.2 | 48.0 |
SimpleQA Verified SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information. | 23.6 | 50.6 |
SWE-Bench verified | 71.3 | 73.5 |
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. | 70.6 | 65.0 |
Terminal Bench Terminal Bench · tests the ability to accomplish real-world tasks using terminal commands, evaluating shell scripting and CLI tool proficiency. | 46.5 | 49.6 |
VPCT VPCT (Visual Pattern Completion Test) · tests visual reasoning and pattern recognition by having models complete visual sequences and transformations. | 9.7 | 49.0 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 47.7 | 60.7 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| Claude Sonnet 4.5 | $3.00 | $15.00 | 1.0M tokens (~500 books) | $60.00 |
| GPT-5 | $1.25 | $10.00 | 400K tokens (~200 books) | $34.38 |