Compare · ModelsLive · 2 picked · head to head
Claude Sonnet 4.6 vs GPT-5
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Claude Sonnet 4.6 wins on 10/17 benchmarks
Claude Sonnet 4.6 wins 10 of 17 shared benchmarks. Leads in agentic · reasoning · coding.
Category leads
agentic·Claude Sonnet 4.6reasoning·Claude Sonnet 4.6knowledge·GPT-5general·GPT-5math·GPT-5coding·Claude Sonnet 4.6
Hype vs Reality
Attention vs performance
Claude Sonnet 4.6
#169 by perf·#11 by attention
GPT-5
#116 by perf·#9 by attention
Best value
GPT-5
1.8x better value than Claude Sonnet 4.6
Claude Sonnet 4.6
5.0 pts/$
$9.00/M
GPT-5
9.3 pts/$
$5.63/M
Vendor risk
Who is behind the model
Anthropic
$965.0B·Tier 1
OpenAI
$840.0B·Tier 1
Head to head
17 benchmarks · 2 models
Claude Sonnet 4.6GPT-5
APEX-Agents
Claude Sonnet 4.6 leads by +24.7
APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments.
Claude Sonnet 4.6
43.0
GPT-5
18.3
ARC-AGI
Claude Sonnet 4.6 leads by +20.8
ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.
Claude Sonnet 4.6
86.5
GPT-5
65.7
ARC-AGI-2
Claude Sonnet 4.6 leads by +50.6
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 4.6
60.4
GPT-5
9.9
Chess Puzzles
GPT-5 leads by +24.0
Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities.
Claude Sonnet 4.6
13.0
GPT-5
37.0
DeepResearch Bench
GPT-5 leads by +0.2
DeepResearch Bench · evaluates AI on complex multi-step research tasks requiring information gathering, synthesis, and producing comprehensive analyses.
Claude Sonnet 4.6
54.9
GPT-5
55.1
Dtbench
GPT-5 leads by +1.3
Claude Sonnet 4.6
83.1
GPT-5
84.5
FrontierMath-2025-02-28-Private
GPT-5 leads by +0.0
FrontierMath (Feb 2025) · original research-level math problems created by mathematicians, testing capabilities at the boundary of current AI mathematical reasoning.
Claude Sonnet 4.6
32.4
GPT-5
32.4
FrontierMath-Tier-4-2025-07-01-Private
Claude Sonnet 4.6 leads by +1.3
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 4.6
13.8
GPT-5
12.5
GPQA diamond
Claude Sonnet 4.6 leads by +1.7
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
Claude Sonnet 4.6
86.6
GPT-5
85.0
Lmca
Claude Sonnet 4.6 leads by +7.7
Claude Sonnet 4.6
54.7
GPT-5
47.0
Mystery Game Puzzles
GPT-5 leads by +7.7
Claude Sonnet 4.6
7.5
GPT-5
15.2
OTIS Mock AIME 2024-2025
GPT-5 leads by +5.6
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Claude Sonnet 4.6
85.8
GPT-5
91.4
Proofbench
Claude Sonnet 4.6 leads by +27.0
Claude Sonnet 4.6
45.0
GPT-5
18.0
SimpleQA Verified
GPT-5 leads by +15.1
SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information.
Claude Sonnet 4.6
35.5
GPT-5
50.6
SWE-Bench verified
Claude Sonnet 4.6 leads by +1.7
SWE-bench Verified · 500 human-validated tasks from 12 real Python repositories (Django, Flask, scikit-learn, sympy, and others). Each task requires the model to produce a git patch that resolves a real GitHub issue and passes the test suite. The verified subset eliminates ambiguous tasks from the original SWE-bench. The benchmark remains the most-cited evaluation for code-generation capability.
Claude Sonnet 4.6
75.2
GPT-5
73.5
Terminal Bench
Claude Sonnet 4.6 leads by +3.8
Terminal-Bench 2.0 · evaluates AI agents on real terminal-based coding tasks · writing scripts, debugging, running tests, and managing projects entirely through command-line interaction. Tests both code quality and terminal fluency.
Claude Sonnet 4.6
53.4
GPT-5
49.6
WeirdML
Claude Sonnet 4.6 leads by +5.4
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
Claude Sonnet 4.6
66.1
GPT-5
60.7
Full benchmark table
| Benchmark | Claude Sonnet 4.6 | GPT-5 |
|---|---|---|
APEX-Agents APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments. | 43.0 | 18.3 |
ARC-AGI ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization. | 86.5 | 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. | 60.4 | 9.9 |
Chess Puzzles Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities. | 13.0 | 37.0 |
DeepResearch Bench DeepResearch Bench · evaluates AI on complex multi-step research tasks requiring information gathering, synthesis, and producing comprehensive analyses. | 54.9 | 55.1 |
Dtbench | 83.1 | 84.5 |
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. | 32.4 | 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. | 13.8 | 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. | 86.6 | 85.0 |
Lmca | 54.7 | 47.0 |
Mystery Game Puzzles | 7.5 | 15.2 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 85.8 | 91.4 |
Proofbench | 45.0 | 18.0 |
SimpleQA Verified SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information. | 35.5 | 50.6 |
SWE-Bench verified SWE-bench Verified · 500 human-validated tasks from 12 real Python repositories (Django, Flask, scikit-learn, sympy, and others). Each task requires the model to produce a git patch that resolves a real GitHub issue and passes the test suite. The verified subset eliminates ambiguous tasks from the original SWE-bench. The benchmark remains the most-cited evaluation for code-generation capability. | 75.2 | 73.5 |
Terminal Bench Terminal-Bench 2.0 · evaluates AI agents on real terminal-based coding tasks · writing scripts, debugging, running tests, and managing projects entirely through command-line interaction. Tests both code quality and terminal fluency. | 53.4 | 49.6 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 66.1 | 60.7 |
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 | |
| $1.25 | $10.00 | 400K tokens (~200 books) | $34.38 |