Compare · ModelsLive · 2 picked · head to head
Claude Haiku 4.5 vs GPT-5 Nano
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Claude Haiku 4.5 wins on 7/10 benchmarks
Claude Haiku 4.5 wins 7 of 10 shared benchmarks. Leads in reasoning · knowledge · coding.
Category leads
reasoning·Claude Haiku 4.5math·GPT-5 Nanoknowledge·Claude Haiku 4.5coding·Claude Haiku 4.5
Hype vs Reality
Attention vs performance
Claude Haiku 4.5
#161 by perf·no signal
GPT-5 Nano
#114 by perf·no signal
Best value
GPT-5 Nano
16.3x better value than Claude Haiku 4.5
Claude Haiku 4.5
12.4 pts/$
$3.00/M
GPT-5 Nano
201.3 pts/$
$0.23/M
Vendor risk
Who is behind the model
Anthropic
$380.0B·Tier 1
OpenAI
$840.0B·Tier 1
Head to head
10 benchmarks · 2 models
Claude Haiku 4.5GPT-5 Nano
ARC-AGI
Claude Haiku 4.5 leads by +27.0
ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.
Claude Haiku 4.5
47.7
GPT-5 Nano
20.7
ARC-AGI-2
Claude Haiku 4.5 leads by +1.4
ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data.
Claude Haiku 4.5
4.0
GPT-5 Nano
2.6
FrontierMath-2025-02-28-Private
GPT-5 Nano leads by +2.4
FrontierMath (Feb 2025) · original research-level math problems created by mathematicians, testing capabilities at the boundary of current AI mathematical reasoning.
Claude Haiku 4.5
5.9
GPT-5 Nano
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.
Claude Haiku 4.5
2.1
GPT-5 Nano
2.1
GPQA diamond
Claude Haiku 4.5 leads by +2.4
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
Claude Haiku 4.5
61.6
GPT-5 Nano
59.3
MATH level 5
Claude Haiku 4.5 leads by +1.1
MATH Level 5 · the hardest tier of the MATH benchmark, featuring competition-level problems from AMC, AIME, and Olympiad-style mathematics.
Claude Haiku 4.5
96.4
GPT-5 Nano
95.2
OTIS Mock AIME 2024-2025
GPT-5 Nano leads by +14.5
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Claude Haiku 4.5
66.6
GPT-5 Nano
81.1
SimpleQA Verified
GPT-5 Nano leads by +6.3
SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information.
Claude Haiku 4.5
5.9
GPT-5 Nano
12.2
Terminal Bench
Claude Haiku 4.5 leads by +24.0
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 Haiku 4.5
35.5
GPT-5 Nano
11.5
WeirdML
Claude Haiku 4.5 leads by +7.3
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
Claude Haiku 4.5
45.4
GPT-5 Nano
38.1
Full benchmark table
| Benchmark | Claude Haiku 4.5 | 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. | 47.7 | 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. | 4.0 | 2.6 |
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. | 5.9 | 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. | 2.1 | 2.1 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 61.6 | 59.3 |
MATH level 5 MATH Level 5 · the hardest tier of the MATH benchmark, featuring competition-level problems from AMC, AIME, and Olympiad-style mathematics. | 96.4 | 95.2 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 66.6 | 81.1 |
SimpleQA Verified SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information. | 5.9 | 12.2 |
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. | 35.5 | 11.5 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 45.4 | 38.1 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| $1.00 | $5.00 | 200K tokens (~100 books) | $20.00 | |
| $0.05 | $0.40 | 400K tokens (~200 books) | $1.38 |