Compare · ModelsLive · 2 picked · head to head
Claude Sonnet 4.5 vs Grok 3
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Claude Sonnet 4.5 wins on 9/9 benchmarks
Claude Sonnet 4.5 wins 9 of 9 shared benchmarks. Leads in reasoning · math · knowledge.
Category leads
reasoning·Claude Sonnet 4.5math·Claude Sonnet 4.5knowledge·Claude Sonnet 4.5coding·Claude Sonnet 4.5
Hype vs Reality
Attention vs performance
Claude Sonnet 4.5
#174 by perf·no signal
Grok 3
#141 by perf·#9 by attention
Vendor risk
Who is behind the model
Anthropic
$380.0B·Tier 1
xAI
$250.0B·Tier 1
Head to head
9 benchmarks · 2 models
Claude Sonnet 4.5Grok 3
ARC-AGI
Claude Sonnet 4.5 leads by +58.2
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
Grok 3
5.5
ARC-AGI-2
Claude Sonnet 4.5 leads by +13.6
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
Grok 3
0.0
FrontierMath-2025-02-28-Private
Claude Sonnet 4.5 leads by +11.4
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
Grok 3
3.8
FrontierMath-Tier-4-2025-07-01-Private
Claude Sonnet 4.5 leads by +4.2
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
Grok 3
0.0
GPQA diamond
Claude Sonnet 4.5 leads by +8.8
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
Grok 3
67.7
MATH level 5
Claude Sonnet 4.5 leads by +9.0
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
Grok 3
88.8
OTIS Mock AIME 2024-2025
Claude Sonnet 4.5 leads by +22.3
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Claude Sonnet 4.5
77.8
Grok 3
55.5
SimpleBench
Claude Sonnet 4.5 leads by +21.8
SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking.
Claude Sonnet 4.5
45.2
Grok 3
23.3
WeirdML
Claude Sonnet 4.5 leads by +10.5
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
Claude Sonnet 4.5
47.7
Grok 3
37.2
Full benchmark table
| Benchmark | Claude Sonnet 4.5 | Grok 3 |
|---|---|---|
ARC-AGI ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization. | 63.7 | 5.5 |
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 | 0.0 |
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 | 3.8 |
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 | 0.0 |
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 | 67.7 |
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 | 88.8 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 77.8 | 55.5 |
SimpleBench SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking. | 45.2 | 23.3 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 47.7 | 37.2 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| $3.00 | $15.00 | 1.0M tokens (~500 books) | $60.00 | |
| — | — | — | — |