Compare · ModelsLive · 2 picked · head to head
Claude Sonnet 4.6 vs GPT-5.2
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
GPT-5.2 wins on 10/15 benchmarks
GPT-5.2 wins 10 of 15 shared benchmarks. Leads in agentic · knowledge · math.
Category leads
agentic·GPT-5.2reasoning·Claude Sonnet 4.6arena·Claude Sonnet 4.6knowledge·GPT-5.2math·GPT-5.2coding·GPT-5.2
Hype vs Reality
Attention vs performance
Claude Sonnet 4.6
#128 by perf·#18 by attention
GPT-5.2
#93 by perf·no signal
Best value
GPT-5.2
1.3x better value than Claude Sonnet 4.6
Claude Sonnet 4.6
5.3 pts/$
$9.00/M
GPT-5.2
6.8 pts/$
$7.88/M
Vendor risk
Who is behind the model
Anthropic
$380.0B·Tier 1
OpenAI
$840.0B·Tier 1
Head to head
15 benchmarks · 2 models
Claude Sonnet 4.6GPT-5.2
APEX-Agents
GPT-5.2 leads by +10.6
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
23.7
GPT-5.2
34.3
ARC-AGI
Claude Sonnet 4.6 leads by +0.3
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-5.2
86.2
ARC-AGI-2
Claude Sonnet 4.6 leads by +7.5
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-5.2
52.9
Chatbot Arena Elo · Coding
Claude Sonnet 4.6 leads by +116.1
Claude Sonnet 4.6
1521.0
GPT-5.2
1405.0
Chatbot Arena Elo · Overall
Claude Sonnet 4.6 leads by +37.1
Claude Sonnet 4.6
1471.7
GPT-5.2
1434.6
Chess Puzzles
GPT-5.2 leads by +36.0
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-5.2
49.0
FrontierMath-2025-02-28-Private
GPT-5.2 leads by +8.3
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-5.2
40.7
FrontierMath-Tier-4-2025-07-01-Private
GPT-5.2 leads by +10.5
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
8.3
GPT-5.2
18.8
GPQA diamond
GPT-5.2 leads by +5.4
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
83.2
GPT-5.2
88.5
OTIS Mock AIME 2024-2025
GPT-5.2 leads by +10.3
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Claude Sonnet 4.6
85.8
GPT-5.2
96.1
PostTrainBench
GPT-5.2 leads by +5.0
Claude Sonnet 4.6
16.4
GPT-5.2
21.4
SimpleQA Verified
GPT-5.2 leads by +9.9
SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information.
Claude Sonnet 4.6
29.0
GPT-5.2
38.9
SWE-Bench verified
Claude Sonnet 4.6 leads by +1.4
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. Claude Mythos Preview leads at 93.9%, crossing 90% for the first time in 2026. Opus 4.6 scores 80.8%. The benchmark remains the most-cited evaluation for code-generation capability.
Claude Sonnet 4.6
75.2
GPT-5.2
73.8
Terminal Bench
GPT-5.2 leads by +11.5
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 Opus 4.7 scores 69.4%, demonstrating significant agentic terminal competence.
Claude Sonnet 4.6
53.4
GPT-5.2
64.9
WeirdML
GPT-5.2 leads by +6.1
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-5.2
72.2
Full benchmark table
| Benchmark | Claude Sonnet 4.6 | GPT-5.2 |
|---|---|---|
APEX-Agents APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments. | 23.7 | 34.3 |
ARC-AGI ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization. | 86.5 | 86.2 |
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 | 52.9 |
Chatbot Arena Elo · Coding | 1521.0 | 1405.0 |
Chatbot Arena Elo · Overall | 1471.7 | 1434.6 |
Chess Puzzles Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities. | 13.0 | 49.0 |
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. | 32.4 | 40.7 |
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. | 8.3 | 18.8 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 83.2 | 88.5 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 85.8 | 96.1 |
PostTrainBench | 16.4 | 21.4 |
SimpleQA Verified SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information. | 29.0 | 38.9 |
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. Claude Mythos Preview leads at 93.9%, crossing 90% for the first time in 2026. Opus 4.6 scores 80.8%. The benchmark remains the most-cited evaluation for code-generation capability. | 75.2 | 73.8 |
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. Claude Opus 4.7 scores 69.4%, demonstrating significant agentic terminal competence. | 53.4 | 64.9 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 66.1 | 72.2 |
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 | |
| $1.75 | $14.00 | 400K tokens (~200 books) | $48.13 |
People also compared