Compare · ModelsLive · 2 picked · head to head
Claude Sonnet 4 vs Llama 4 Maverick
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Claude Sonnet 4 wins on 12/13 benchmarks
Claude Sonnet 4 wins 12 of 13 shared benchmarks. Leads in coding · reasoning · knowledge.
Category leads
coding·Claude Sonnet 4reasoning·Claude Sonnet 4knowledge·Claude Sonnet 4math·Claude Sonnet 4
Hype vs Reality
Attention vs performance
Claude Sonnet 4
#115 by perf·no signal
Llama 4 Maverick
#193 by perf·no signal
Best value
Llama 4 Maverick
15.1x better value than Claude Sonnet 4
Claude Sonnet 4
5.0 pts/$
$9.00/M
Llama 4 Maverick
74.7 pts/$
$0.38/M
Vendor risk
Who is behind the model
Anthropic
$380.0B·Tier 1
Meta AI
$1.50T·Tier 1
Head to head
13 benchmarks · 2 models
Claude Sonnet 4Llama 4 Maverick
Aider polyglot
Claude Sonnet 4 leads by +45.7
Aider Polyglot · measures how well AI models can edit code across multiple programming languages using the Aider coding assistant framework.
Claude Sonnet 4
61.3
Llama 4 Maverick
15.6
ARC-AGI
Claude Sonnet 4 leads by +35.6
ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.
Claude Sonnet 4
40.0
Llama 4 Maverick
4.4
ARC-AGI-2
Claude Sonnet 4 leads by +5.8
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
5.9
Llama 4 Maverick
0.1
Fiction.LiveBench
Claude Sonnet 4 leads by +0.7
Fiction.LiveBench · a continuously updated benchmark using recently published fiction to test reading comprehension and reasoning, preventing data contamination.
Claude Sonnet 4
46.9
Llama 4 Maverick
46.2
FrontierMath-2025-02-28-Private
Claude Sonnet 4 leads by +3.4
FrontierMath (Feb 2025) · original research-level math problems created by mathematicians, testing capabilities at the boundary of current AI mathematical reasoning.
Claude Sonnet 4
4.1
Llama 4 Maverick
0.7
GeoBench
Llama 4 Maverick leads by +15.0
GeoBench · tests geographic knowledge and spatial reasoning across countries, landmarks, coordinates, and geopolitical understanding.
Claude Sonnet 4
37.0
Llama 4 Maverick
52.0
GPQA diamond
Claude Sonnet 4 leads by +16.3
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
72.3
Llama 4 Maverick
56.0
HLE
Claude Sonnet 4 leads by +2.2
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
3.1
Llama 4 Maverick
0.9
MATH level 5
Claude Sonnet 4 leads by +11.4
MATH Level 5 · the hardest tier of the MATH benchmark, featuring competition-level problems from AMC, AIME, and Olympiad-style mathematics.
Claude Sonnet 4
84.4
Llama 4 Maverick
73.0
OTIS Mock AIME 2024-2025
Claude Sonnet 4 leads by +50.6
OTIS Mock AIME 2024–2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Claude Sonnet 4
71.1
Llama 4 Maverick
20.5
SimpleBench
Claude Sonnet 4 leads by +21.4
SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking.
Claude Sonnet 4
34.6
Llama 4 Maverick
13.2
SWE-Bench Verified (Bash Only)
Claude Sonnet 4 leads by +43.9
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
64.9
Llama 4 Maverick
21.0
WeirdML
Claude Sonnet 4 leads by +21.6
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
Claude Sonnet 4
46.1
Llama 4 Maverick
24.5
Full benchmark table
| Benchmark | Claude Sonnet 4 | Llama 4 Maverick |
|---|---|---|
Aider polyglot Aider Polyglot · measures how well AI models can edit code across multiple programming languages using the Aider coding assistant framework. | 61.3 | 15.6 |
ARC-AGI ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization. | 40.0 | 4.4 |
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. | 5.9 | 0.1 |
Fiction.LiveBench Fiction.LiveBench · a continuously updated benchmark using recently published fiction to test reading comprehension and reasoning, preventing data contamination. | 46.9 | 46.2 |
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. | 4.1 | 0.7 |
GeoBench GeoBench · tests geographic knowledge and spatial reasoning across countries, landmarks, coordinates, and geopolitical understanding. | 37.0 | 52.0 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 72.3 | 56.0 |
HLE HLE (Humanity's Last Exam) · crowdsourced expert-level questions designed to be among the hardest possible challenges for AI systems across all domains. | 3.1 | 0.9 |
MATH level 5 MATH Level 5 · the hardest tier of the MATH benchmark, featuring competition-level problems from AMC, AIME, and Olympiad-style mathematics. | 84.4 | 73.0 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024–2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 71.1 | 20.5 |
SimpleBench SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking. | 34.6 | 13.2 |
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. | 64.9 | 21.0 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 46.1 | 24.5 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| Claude Sonnet 4 | $3.00 | $15.00 | 1.0M tokens (~500 books) | $60.00 |
| Llama 4 Maverick | $0.15 | $0.60 | 1.0M tokens (~524 books) | $2.62 |