Compare · ModelsLive · 2 picked · head to head
GPT-4.1 vs GPT-5 Nano
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
GPT-5 Nano wins on 12/20 benchmarks
GPT-5 Nano wins 12 of 20 shared benchmarks. Leads in reasoning · knowledge · math.
Category leads
reasoning·GPT-5 Nanoknowledge·GPT-5 Nanogeneral·GPT-4.1math·GPT-5 Nanolanguage·GPT-5 Nanocoding·GPT-4.1
Hype vs Reality
Attention vs performance
GPT-4.1
#189 by perf·no signal
GPT-5 Nano
#193 by perf·#4 by attention
Best value
GPT-5 Nano
22.1x better value than GPT-4.1
GPT-4.1
7.9 pts/$
$5.00/M
GPT-5 Nano
174.2 pts/$
$0.23/M
Vendor risk
Who is behind the model
OpenAI
$840.0B·Tier 1
OpenAI
$840.0B·Tier 1
Head to head
20 benchmarks · 2 models
GPT-4.1GPT-5 Nano
ARC-AGI
GPT-5 Nano leads by +15.2
ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.
GPT-4.1
5.5
GPT-5 Nano
20.7
ARC-AGI-2
GPT-5 Nano leads by +2.2
ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data.
GPT-4.1
0.4
GPT-5 Nano
2.6
Chess Puzzles
GPT-5 Nano leads by +22.1
Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities.
GPT-4.1
1.1
GPT-5 Nano
23.2
Dtbench
GPT-4.1 leads by +9.3
GPT-4.1
47.1
GPT-5 Nano
37.8
Fiction.LiveBench
GPT-4.1 leads by +19.5
Fiction.LiveBench · a continuously updated benchmark using recently published fiction to test reading comprehension and reasoning, preventing data contamination.
GPT-4.1
63.9
GPT-5 Nano
44.4
FrontierMath-2025-02-28-Private
GPT-5 Nano leads by +2.8
FrontierMath (Feb 2025) · original research-level math problems created by mathematicians, testing capabilities at the boundary of current AI mathematical reasoning.
GPT-4.1
5.5
GPT-5 Nano
8.3
FrontierMath-Tier-4-2025-07-01-Private
GPT-5 Nano leads by +2.1
FrontierMath Tier 4 (Jul 2025) · the most challenging tier of frontier mathematics, containing problems that push the absolute limits of AI mathematical reasoning.
GPT-4.1
0.0
GPT-5 Nano
2.1
FrontierMath-Tiers-1-3-v2-Private
GPT-5 Nano leads by +14.0
GPT-4.1
6.0
GPT-5 Nano
20.0
GPQA diamond
GPT-5 Nano leads by +3.4
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
GPT-4.1
55.9
GPT-5 Nano
59.3
HELM · GPQA
GPT-5 Nano leads by +2.0
GPT-4.1
65.9
GPT-5 Nano
67.9
HELM · IFEval
GPT-5 Nano leads by +9.4
GPT-4.1
83.8
GPT-5 Nano
93.2
HELM · MMLU-Pro
GPT-4.1 leads by +3.3
GPT-4.1
81.1
GPT-5 Nano
77.8
HELM · Omni-MATH
GPT-5 Nano leads by +7.6
GPT-4.1
47.1
GPT-5 Nano
54.7
HELM · WildBench
GPT-4.1 leads by +4.8
GPT-4.1
85.4
GPT-5 Nano
80.6
Lmca
GPT-4.1 leads by +20.8
GPT-4.1
30.1
GPT-5 Nano
9.3
MATH level 5
GPT-5 Nano leads by +12.2
MATH Level 5 · the hardest tier of the MATH benchmark, featuring competition-level problems from AMC, AIME, and Olympiad-style mathematics.
GPT-4.1
83.0
GPT-5 Nano
95.2
OTIS Mock AIME 2024-2025
GPT-5 Nano leads by +42.8
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
GPT-4.1
38.3
GPT-5 Nano
81.1
SimpleQA Verified
GPT-4.1 leads by +19.4
SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information.
GPT-4.1
31.1
GPT-5 Nano
11.7
SWE-Bench Verified (Bash Only)
GPT-4.1 leads by +4.8
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.
GPT-4.1
39.6
GPT-5 Nano
34.8
WeirdML
GPT-4.1 leads by +1.0
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
GPT-4.1
39.0
GPT-5 Nano
38.1
Full benchmark table
| Benchmark | GPT-4.1 | GPT-5 Nano |
|---|---|---|
ARC-AGI ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization. | 5.5 | 20.7 |
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. | 0.4 | 2.6 |
Chess Puzzles Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities. | 1.1 | 23.2 |
Dtbench | 47.1 | 37.8 |
Fiction.LiveBench Fiction.LiveBench · a continuously updated benchmark using recently published fiction to test reading comprehension and reasoning, preventing data contamination. | 63.9 | 44.4 |
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. | 5.5 | 8.3 |
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. | 0.0 | 2.1 |
FrontierMath-Tiers-1-3-v2-Private | 6.0 | 20.0 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 55.9 | 59.3 |
HELM · GPQA | 65.9 | 67.9 |
HELM · IFEval | 83.8 | 93.2 |
HELM · MMLU-Pro | 81.1 | 77.8 |
HELM · Omni-MATH | 47.1 | 54.7 |
HELM · WildBench | 85.4 | 80.6 |
Lmca | 30.1 | 9.3 |
MATH level 5 MATH Level 5 · the hardest tier of the MATH benchmark, featuring competition-level problems from AMC, AIME, and Olympiad-style mathematics. | 83.0 | 95.2 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 38.3 | 81.1 |
SimpleQA Verified SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information. | 31.1 | 11.7 |
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. | 39.6 | 34.8 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 39.0 | 38.1 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| $2.00 | $8.00 | 1.0M tokens (~524 books) | $35.00 | |
| $0.05 | $0.40 | 400K tokens (~200 books) | $1.38 |