Compare · ModelsLive · 3 picked · head to head
Claude 3.5 Sonnet vs o1 vs o1-preview
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
o1 wins on 12/18 benchmarks
o1 wins 12 of 18 shared benchmarks. Leads in knowledge · math · general.
Category leads
coding·Claude 3.5 Sonnetknowledge·o1math·o1general·o1reasoning·o1-previewarena·o1-preview
Hype vs Reality
Attention vs performance
Claude 3.5 Sonnet
#190 by perf·no signal
o1
#158 by perf·no signal
o1-preview
#174 by perf·no signal
Vendor risk
Who is behind the model
Anthropic
$965.0B·Tier 1
OpenAI
$840.0B·Tier 1
OpenAI
$840.0B·Tier 1
Head to head
18 benchmarks · 3 models
Claude 3.5 Sonneto1o1-preview
Aider · Code Editing
Claude 3.5 Sonnet
84.2
o1
84.2
o1-preview
79.7
GPQA diamond
o1 leads by +30.3
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
Claude 3.5 Sonnet
38.7
o1
69.0
o1-preview
33.8
MATH level 5
o1 leads by +13.1
MATH Level 5 · the hardest tier of the MATH benchmark, featuring competition-level problems from AMC, AIME, and Olympiad-style mathematics.
Claude 3.5 Sonnet
51.7
o1
94.7
o1-preview
81.7
Metr Time Horizons
o1 leads by +5.9
Claude 3.5 Sonnet
40.1
o1
51.0
o1-preview
45.1
OTIS Mock AIME 2024-2025
o1 leads by +42.3
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Claude 3.5 Sonnet
6.4
o1
73.3
o1-preview
31.0
SimpleBench
o1-preview leads by +1.9
SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking.
Claude 3.5 Sonnet
13.0
o1
28.1
o1-preview
30.0
WeirdML
o1-preview leads by +1.5
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
Claude 3.5 Sonnet
31.0
o1
46.1
o1-preview
47.6
Aider polyglot
o1 leads by +10.1
Aider Polyglot · measures how well AI models can edit code across multiple programming languages using the Aider coding assistant framework.
Claude 3.5 Sonnet
51.6
o1
61.7
ARC-AGI
o1 leads by +12.7
ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.
o1
30.7
o1-preview
18.0
Chatbot Arena Elo · Overall
o1-preview leads by +14.5
Claude 3.5 Sonnet
1374.0
o1-preview
1388.5
CadEval
o1 leads by +8.0
CadEval · evaluates the ability to generate and reason about Computer-Aided Design code, testing spatial reasoning and engineering knowledge.
Claude 3.5 Sonnet
48.0
o1
56.0
Cybench
Claude 3.5 Sonnet leads by +7.5
Cybench · evaluates AI on real Capture-The-Flag cybersecurity challenges, testing vulnerability analysis, exploitation, and security reasoning.
Claude 3.5 Sonnet
17.5
o1-preview
10.0
Dtbench
o1 leads by +11.5
Claude 3.5 Sonnet
46.3
o1
57.8
FrontierMath-2025-02-28-Private
o1 leads by +7.5
FrontierMath (Feb 2025) · original research-level math problems created by mathematicians, testing capabilities at the boundary of current AI mathematical reasoning.
Claude 3.5 Sonnet
1.8
o1
9.3
GeoBench
o1 leads by +18.0
GeoBench · tests geographic knowledge and spatial reasoning across countries, landmarks, coordinates, and geopolitical understanding.
Claude 3.5 Sonnet
62.0
o1
80.0
HLE
o1 leads by +3.3
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. Claude Opus 4.7 scores 46.9% without tools and 54.7% with tools · making it one of the few benchmarks where the top score is below 60%.
Claude 3.5 Sonnet
0.0
o1
3.3
Lech Mazur Writing
Claude 3.5 Sonnet leads by +10.1
Lech Mazur Writing · evaluates creative writing ability, assessing prose quality, narrative coherence, and stylistic sophistication.
Claude 3.5 Sonnet
80.3
o1
70.2
VPCT
o1 leads by +5.5
VPCT (Visual Pattern Completion Test) · tests visual reasoning and pattern recognition by having models complete visual sequences and transformations.
Claude 3.5 Sonnet
0.0
o1
5.5
Full benchmark table
| Benchmark | Claude 3.5 Sonnet | o1 | o1-preview |
|---|---|---|---|
Aider · Code Editing | 84.2 | 84.2 | 79.7 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 38.7 | 69.0 | 33.8 |
MATH level 5 MATH Level 5 · the hardest tier of the MATH benchmark, featuring competition-level problems from AMC, AIME, and Olympiad-style mathematics. | 51.7 | 94.7 | 81.7 |
Metr Time Horizons | 40.1 | 51.0 | 45.1 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 6.4 | 73.3 | 31.0 |
SimpleBench SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking. | 13.0 | 28.1 | 30.0 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 31.0 | 46.1 | 47.6 |
Aider polyglot Aider Polyglot · measures how well AI models can edit code across multiple programming languages using the Aider coding assistant framework. | 51.6 | 61.7 | — |
ARC-AGI ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization. | — | 30.7 | 18.0 |
Chatbot Arena Elo · Overall | 1374.0 | — | 1388.5 |
CadEval CadEval · evaluates the ability to generate and reason about Computer-Aided Design code, testing spatial reasoning and engineering knowledge. | 48.0 | 56.0 | — |
Cybench Cybench · evaluates AI on real Capture-The-Flag cybersecurity challenges, testing vulnerability analysis, exploitation, and security reasoning. | 17.5 | — | 10.0 |
Dtbench | 46.3 | 57.8 | — |
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. | 1.8 | 9.3 | — |
GeoBench GeoBench · tests geographic knowledge and spatial reasoning across countries, landmarks, coordinates, and geopolitical understanding. | 62.0 | 80.0 | — |
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. Claude Opus 4.7 scores 46.9% without tools and 54.7% with tools · making it one of the few benchmarks where the top score is below 60%. | 0.0 | 3.3 | — |
Lech Mazur Writing Lech Mazur Writing · evaluates creative writing ability, assessing prose quality, narrative coherence, and stylistic sophistication. | 80.3 | 70.2 | — |
VPCT VPCT (Visual Pattern Completion Test) · tests visual reasoning and pattern recognition by having models complete visual sequences and transformations. | 0.0 | 5.5 | — |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| — | — | — | — | |
| $15.00 | $60.00 | 200K tokens (~100 books) | $262.50 | |
| — | — | — | — |