Compare · ModelsLive · 2 picked · head to head
Claude Opus 4.6 vs Claude Sonnet 4
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Claude Opus 4.6 wins on 17/17 benchmarks
Claude Opus 4.6 wins 17 of 17 shared benchmarks. Leads in agentic · reasoning · coding.
Category leads
agentic·Claude Opus 4.6reasoning·Claude Opus 4.6coding·Claude Opus 4.6knowledge·Claude Opus 4.6general·Claude Opus 4.6math·Claude Opus 4.6safety·Claude Opus 4.6
Hype vs Reality
Attention vs performance
Claude Opus 4.6
#136 by perf·#9 by attention
Claude Sonnet 4
#157 by perf·#10 by attention
Best value
Claude Sonnet 4
1.6x better value than Claude Opus 4.6
Claude Opus 4.6
3.2 pts/$
$15.00/M
Claude Sonnet 4
5.1 pts/$
$9.00/M
Vendor risk
Who is behind the model
Anthropic
$965.0B·Tier 1
Anthropic
$965.0B·Tier 1
Head to head
17 benchmarks · 2 models
Claude Opus 4.6Claude Sonnet 4
APEX-Agents
Claude Opus 4.6 leads by +37.0
APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments.
Claude Opus 4.6
46.3
Claude Sonnet 4
9.3
ARC-AGI
Claude Opus 4.6 leads by +54.0
ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.
Claude Opus 4.6
94.0
Claude Sonnet 4
40.0
ARC-AGI-2
Claude Opus 4.6 leads by +63.2
ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data.
Claude Opus 4.6
69.2
Claude Sonnet 4
5.9
Cybench
Claude Opus 4.6 leads by +58.0
Cybench · evaluates AI on real Capture-The-Flag cybersecurity challenges, testing vulnerability analysis, exploitation, and security reasoning.
Claude Opus 4.6
93.0
Claude Sonnet 4
35.0
DeepResearch Bench
Claude Opus 4.6 leads by +7.5
DeepResearch Bench · evaluates AI on complex multi-step research tasks requiring information gathering, synthesis, and producing comprehensive analyses.
Claude Opus 4.6
55.3
Claude Sonnet 4
47.8
Dtbench
Claude Opus 4.6 leads by +23.5
Claude Opus 4.6
85.3
Claude Sonnet 4
61.8
FrontierMath-2025-02-28-Private
Claude Opus 4.6 leads by +33.4
FrontierMath (Feb 2025) · original research-level math problems created by mathematicians, testing capabilities at the boundary of current AI mathematical reasoning.
Claude Opus 4.6
40.7
Claude Sonnet 4
7.3
FrontierMath-Tier-4-2025-07-01-Private
Claude Opus 4.6 leads by +22.9
FrontierMath Tier 4 (Jul 2025) · the most challenging tier of frontier mathematics, containing problems that push the absolute limits of AI mathematical reasoning.
Claude Opus 4.6
22.9
Claude Sonnet 4
0.0
GPQA diamond
Claude Opus 4.6 leads by +15.1
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
Claude Opus 4.6
87.4
Claude Sonnet 4
72.3
GSO-Bench
Claude Opus 4.6 leads by +36.3
GSO-Bench · evaluates AI models on real-world open-source software engineering tasks, testing the ability to understand and resolve actual GitHub issues.
Claude Opus 4.6
41.2
Claude Sonnet 4
4.9
HLE
Claude Opus 4.6 leads by +28.0
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 Opus 4.6
31.1
Claude Sonnet 4
3.1
Lmca
Claude Opus 4.6 leads by +31.5
Claude Opus 4.6
65.6
Claude Sonnet 4
34.1
Metr Time Horizons
Claude Opus 4.6 leads by +16.9
Claude Opus 4.6
78.9
Claude Sonnet 4
62.0
OTIS Mock AIME 2024-2025
Claude Opus 4.6 leads by +23.4
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Claude Opus 4.6
94.4
Claude Sonnet 4
71.1
MASK
Claude Opus 4.6 leads by +1.0
Claude Opus 4.6
96.3
Claude Sonnet 4
95.3
SimpleBench
Claude Opus 4.6 leads by +26.5
SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking.
Claude Opus 4.6
61.1
Claude Sonnet 4
34.6
WeirdML
Claude Opus 4.6 leads by +31.8
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
Claude Opus 4.6
78.0
Claude Sonnet 4
46.1
Full benchmark table
| Benchmark | Claude Opus 4.6 | Claude Sonnet 4 |
|---|---|---|
APEX-Agents APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments. | 46.3 | 9.3 |
ARC-AGI ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization. | 94.0 | 40.0 |
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. | 69.2 | 5.9 |
Cybench Cybench · evaluates AI on real Capture-The-Flag cybersecurity challenges, testing vulnerability analysis, exploitation, and security reasoning. | 93.0 | 35.0 |
DeepResearch Bench DeepResearch Bench · evaluates AI on complex multi-step research tasks requiring information gathering, synthesis, and producing comprehensive analyses. | 55.3 | 47.8 |
Dtbench | 85.3 | 61.8 |
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. | 40.7 | 7.3 |
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. | 22.9 | 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. | 87.4 | 72.3 |
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. | 41.2 | 4.9 |
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%. | 31.1 | 3.1 |
Lmca | 65.6 | 34.1 |
Metr Time Horizons | 78.9 | 62.0 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 94.4 | 71.1 |
MASK | 96.3 | 95.3 |
SimpleBench SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking. | 61.1 | 34.6 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 78.0 | 46.1 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| $5.00 | $25.00 | 1.0M tokens (~500 books) | $100.00 | |
| $3.00 | $15.00 | 200K tokens (~100 books) | $60.00 |