Compare · ModelsLive · 2 picked · head to head
Gemini 3.8 Flash vs Gemini 3 Pro
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Gemini 3.8 Flash wins on 10/12 benchmarks
Gemini 3.8 Flash wins 10 of 12 shared benchmarks. Leads in agentic · reasoning · knowledge.
Category leads
speed·Gemini 3 Proagentic·Gemini 3.8 Flashreasoning·Gemini 3.8 Flashknowledge·Gemini 3.8 Flashmath·Gemini 3.8 Flashgeneral·Gemini 3.8 Flashcoding·Gemini 3.8 Flash
Hype vs Reality
Attention vs performance
Gemini 3.8 Flash
#53 by perf·#6 by attention
Gemini 3 Pro
#73 by perf·#6 by attention
Vendor risk
Who is behind the model
Google DeepMind
$4.27T·Tier 1
Google DeepMind
$4.27T·Tier 1
Head to head
12 benchmarks · 2 models
Gemini 3.8 FlashGemini 3 Pro
Artificial Analysis · Quality Index
Gemini 3 Pro leads by +0.4
Gemini 3.8 Flash
40.9
Gemini 3 Pro
41.3
APEX-Agents
Gemini 3.8 Flash leads by +45.9
APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments.
Gemini 3.8 Flash
64.3
Gemini 3 Pro
18.4
ARC-AGI
Gemini 3.8 Flash leads by +23.5
ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.
Gemini 3.8 Flash
98.5
Gemini 3 Pro
75.0
ARC-AGI-2
Gemini 3.8 Flash leads by +58.1
ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data.
Gemini 3.8 Flash
89.2
Gemini 3 Pro
31.1
Chess Puzzles
Gemini 3.8 Flash leads by +28.4
Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities.
Gemini 3.8 Flash
55.8
Gemini 3 Pro
27.4
GPQA diamond
Gemini 3.8 Flash leads by +3.7
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
Gemini 3.8 Flash
93.9
Gemini 3 Pro
90.2
HLE
Gemini 3.8 Flash leads by +7.4
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%.
Gemini 3.8 Flash
41.7
Gemini 3 Pro
34.4
OTIS Mock AIME 2024-2025
Gemini 3.8 Flash leads by +7.5
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Gemini 3.8 Flash
98.9
Gemini 3 Pro
91.4
Proofbench
Gemini 3.8 Flash leads by +28.0
Gemini 3.8 Flash
48.0
Gemini 3 Pro
20.0
Remote Labor Index
Gemini 3.8 Flash leads by +4.6
Gemini 3.8 Flash
5.8
Gemini 3 Pro
1.3
SimpleQA Verified
Gemini 3 Pro leads by +3.2
SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information.
Gemini 3.8 Flash
69.7
Gemini 3 Pro
72.9
WeirdML
Gemini 3.8 Flash leads by +14.9
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
Gemini 3.8 Flash
84.8
Gemini 3 Pro
69.9
Full benchmark table
| Benchmark | Gemini 3.8 Flash | Gemini 3 Pro |
|---|---|---|
Artificial Analysis · Quality Index | 40.9 | 41.3 |
APEX-Agents APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments. | 64.3 | 18.4 |
ARC-AGI ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization. | 98.5 | 75.0 |
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. | 89.2 | 31.1 |
Chess Puzzles Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities. | 55.8 | 27.4 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 93.9 | 90.2 |
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%. | 41.7 | 34.4 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 98.9 | 91.4 |
Proofbench | 48.0 | 20.0 |
Remote Labor Index | 5.8 | 1.3 |
SimpleQA Verified SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information. | 69.7 | 72.9 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 84.8 | 69.9 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| $0.75 | $3.75 | 1.0M tokens (~524 books) | $15.00 | |
| — | — | — | — |