Compare · ModelsLive · 2 picked · head to head
Gemini 2.5 Flash vs GPT-5 Nano
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Gemini 2.5 Flash wins on 9/16 benchmarks
Gemini 2.5 Flash wins 9 of 16 shared benchmarks. Leads in reasoning · general · knowledge.
Category leads
reasoning·Gemini 2.5 Flashgeneral·Gemini 2.5 Flashknowledge·Gemini 2.5 Flashmath·Gemini 2.5 Flashlanguage·GPT-5 Nanocoding·GPT-5 Nano
Hype vs Reality
Attention vs performance
Gemini 2.5 Flash
#198 by perf·no signal
GPT-5 Nano
#203 by perf·#10 by attention
Best value
GPT-5 Nano
6.2x better value than Gemini 2.5 Flash
Gemini 2.5 Flash
28.1 pts/$
$1.40/M
GPT-5 Nano
173.3 pts/$
$0.23/M
Vendor risk
Who is behind the model
Google DeepMind
$4.27T·Tier 1
OpenAI
$840.0B·Tier 1
Head to head
16 benchmarks · 2 models
Gemini 2.5 FlashGPT-5 Nano
ARC-AGI
Gemini 2.5 Flash leads by +11.6
ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.
Gemini 2.5 Flash
32.3
GPT-5 Nano
20.7
ARC-AGI-2
GPT-5 Nano leads by +0.1
ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data.
Gemini 2.5 Flash
2.5
GPT-5 Nano
2.6
Dtbench
Gemini 2.5 Flash leads by +23.1
Gemini 2.5 Flash
60.9
GPT-5 Nano
37.8
Fiction.LiveBench
Gemini 2.5 Flash leads by +2.8
Fiction.LiveBench · a continuously updated benchmark using recently published fiction to test reading comprehension and reasoning, preventing data contamination.
Gemini 2.5 Flash
47.2
GPT-5 Nano
44.4
FrontierMath-2025-02-28-Private
Gemini 2.5 Flash leads by +0.2
FrontierMath (Feb 2025) · original research-level math problems created by mathematicians, testing capabilities at the boundary of current AI mathematical reasoning.
Gemini 2.5 Flash
8.5
GPT-5 Nano
8.3
FrontierMath-Tier-4-2025-07-01-Private
Gemini 2.5 Flash leads by +4.9
FrontierMath Tier 4 (Jul 2025) · the most challenging tier of frontier mathematics, containing problems that push the absolute limits of AI mathematical reasoning.
Gemini 2.5 Flash
6.9
GPT-5 Nano
2.1
HELM · GPQA
GPT-5 Nano leads by +28.9
Gemini 2.5 Flash
39.0
GPT-5 Nano
67.9
HELM · IFEval
GPT-5 Nano leads by +3.4
Gemini 2.5 Flash
89.8
GPT-5 Nano
93.2
HELM · MMLU-Pro
GPT-5 Nano leads by +13.9
Gemini 2.5 Flash
63.9
GPT-5 Nano
77.8
HELM · Omni-MATH
GPT-5 Nano leads by +16.3
Gemini 2.5 Flash
38.4
GPT-5 Nano
54.7
HELM · WildBench
Gemini 2.5 Flash leads by +1.1
Gemini 2.5 Flash
81.7
GPT-5 Nano
80.6
Lmca
Gemini 2.5 Flash leads by +23.0
Gemini 2.5 Flash
32.3
GPT-5 Nano
9.3
OTIS Mock AIME 2024-2025
GPT-5 Nano leads by +8.1
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Gemini 2.5 Flash
73.0
GPT-5 Nano
81.1
Terminal Bench
GPT-5 Nano leads by +4.7
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.
Gemini 2.5 Flash
17.1
GPT-5 Nano
21.8
VPCT
Gemini 2.5 Flash leads by +1.2
VPCT (Visual Pattern Completion Test) · tests visual reasoning and pattern recognition by having models complete visual sequences and transformations.
Gemini 2.5 Flash
7.0
GPT-5 Nano
5.8
WeirdML
Gemini 2.5 Flash leads by +2.9
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
Gemini 2.5 Flash
41.0
GPT-5 Nano
38.1
Full benchmark table
| Benchmark | Gemini 2.5 Flash | 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. | 32.3 | 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. | 2.5 | 2.6 |
Dtbench | 60.9 | 37.8 |
Fiction.LiveBench Fiction.LiveBench · a continuously updated benchmark using recently published fiction to test reading comprehension and reasoning, preventing data contamination. | 47.2 | 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. | 8.5 | 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. | 6.9 | 2.1 |
HELM · GPQA | 39.0 | 67.9 |
HELM · IFEval | 89.8 | 93.2 |
HELM · MMLU-Pro | 63.9 | 77.8 |
HELM · Omni-MATH | 38.4 | 54.7 |
HELM · WildBench | 81.7 | 80.6 |
Lmca | 32.3 | 9.3 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 73.0 | 81.1 |
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. | 17.1 | 21.8 |
VPCT VPCT (Visual Pattern Completion Test) · tests visual reasoning and pattern recognition by having models complete visual sequences and transformations. | 7.0 | 5.8 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 41.0 | 38.1 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| $0.30 | $2.50 | 1.0M tokens (~524 books) | $8.50 | |
| $0.05 | $0.40 | 400K tokens (~200 books) | $1.38 |