Compare · ModelsLive · 2 picked · head to head
o4 Mini vs GPT-5.1
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
GPT-5.1 wins on 16/19 benchmarks
GPT-5.1 wins 16 of 19 shared benchmarks. Leads in reasoning · knowledge · math.
Category leads
reasoning·GPT-5.1knowledge·GPT-5.1math·GPT-5.1coding·GPT-5.1language·GPT-5.1
Hype vs Reality
Attention vs performance
o4 Mini
#81 by perf·#13 by attention
GPT-5.1
#97 by perf·no signal
Best value
o4 Mini
2.2x better value than GPT-5.1
o4 Mini
19.3 pts/$
$2.75/M
GPT-5.1
8.8 pts/$
$5.63/M
Vendor risk
Who is behind the model
OpenAI
$840.0B·Tier 1
OpenAI
$840.0B·Tier 1
Head to head
19 benchmarks · 2 models
o4 MiniGPT-5.1
ARC-AGI
GPT-5.1 leads by +14.1
ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.
o4 Mini
58.7
GPT-5.1
72.8
ARC-AGI-2
GPT-5.1 leads by +11.5
ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data.
o4 Mini
6.1
GPT-5.1
17.6
Chess Puzzles
GPT-5.1 leads by +6.0
Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities.
o4 Mini
26.0
GPT-5.1
32.0
FrontierMath-2025-02-28-Private
GPT-5.1 leads by +6.2
FrontierMath (Feb 2025) · original research-level math problems created by mathematicians, testing capabilities at the boundary of current AI mathematical reasoning.
o4 Mini
24.8
GPT-5.1
31.0
FrontierMath-Tier-4-2025-07-01-Private
GPT-5.1 leads by +6.3
FrontierMath Tier 4 (Jul 2025) · the most challenging tier of frontier mathematics, containing problems that push the absolute limits of AI mathematical reasoning.
o4 Mini
6.3
GPT-5.1
12.5
GPQA diamond
GPT-5.1 leads by +10.7
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
o4 Mini
72.8
GPT-5.1
83.5
GSO-Bench
GPT-5.1 leads by +10.1
GSO-Bench · evaluates AI models on real-world open-source software engineering tasks, testing the ability to understand and resolve actual GitHub issues.
o4 Mini
3.6
GPT-5.1
13.7
HELM · GPQA
o4 Mini leads by +29.3
o4 Mini
73.5
GPT-5.1
44.2
HELM · IFEval
GPT-5.1 leads by +0.6
o4 Mini
92.9
GPT-5.1
93.5
HELM · MMLU-Pro
o4 Mini leads by +24.1
o4 Mini
82.0
GPT-5.1
57.9
HELM · Omni-MATH
o4 Mini leads by +25.6
o4 Mini
72.0
GPT-5.1
46.4
HELM · WildBench
GPT-5.1 leads by +0.9
o4 Mini
85.4
GPT-5.1
86.3
HLE
GPT-5.1 leads by +5.9
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%.
o4 Mini
13.9
GPT-5.1
19.8
OTIS Mock AIME 2024-2025
GPT-5.1 leads by +6.9
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
o4 Mini
81.7
GPT-5.1
88.6
SimpleBench
GPT-5.1 leads by +17.4
SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking.
o4 Mini
26.4
GPT-5.1
43.8
SimpleQA Verified
GPT-5.1 leads by +25.0
SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information.
o4 Mini
23.9
GPT-5.1
48.9
SWE-Bench Verified (Bash Only)
GPT-5.1 leads by +21.0
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.
o4 Mini
45.0
GPT-5.1
66.0
VPCT
GPT-5.1 leads by +1.8
VPCT (Visual Pattern Completion Test) · tests visual reasoning and pattern recognition by having models complete visual sequences and transformations.
o4 Mini
36.3
GPT-5.1
38.0
WeirdML
GPT-5.1 leads by +8.2
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
o4 Mini
52.6
GPT-5.1
60.8
Full benchmark table
| Benchmark | o4 Mini | GPT-5.1 |
|---|---|---|
ARC-AGI ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization. | 58.7 | 72.8 |
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. | 6.1 | 17.6 |
Chess Puzzles Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities. | 26.0 | 32.0 |
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. | 24.8 | 31.0 |
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. | 6.3 | 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. | 72.8 | 83.5 |
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. | 3.6 | 13.7 |
HELM · GPQA | 73.5 | 44.2 |
HELM · IFEval | 92.9 | 93.5 |
HELM · MMLU-Pro | 82.0 | 57.9 |
HELM · Omni-MATH | 72.0 | 46.4 |
HELM · WildBench | 85.4 | 86.3 |
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%. | 13.9 | 19.8 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 81.7 | 88.6 |
SimpleBench SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking. | 26.4 | 43.8 |
SimpleQA Verified SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information. | 23.9 | 48.9 |
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. | 45.0 | 66.0 |
VPCT VPCT (Visual Pattern Completion Test) · tests visual reasoning and pattern recognition by having models complete visual sequences and transformations. | 36.3 | 38.0 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 52.6 | 60.8 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
People also compared