Compare · ModelsLive · 2 picked · head to head
Claude Opus 4.5 vs Gemini 3.1 Pro Preview
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Gemini 3.1 Pro Preview wins on 23/31 benchmarks
Gemini 3.1 Pro Preview wins 23 of 31 shared benchmarks. Leads in agentic · reasoning · knowledge.
Category leads
agentic·Gemini 3.1 Pro Previewreasoning·Gemini 3.1 Pro Previewarena·Claude Opus 4.5knowledge·Gemini 3.1 Pro Previewgeneral·Claude Opus 4.5math·Gemini 3.1 Pro Previewcoding·Claude Opus 4.5
Head to head · from the data · data as of 2026-11-17
- Benchmarks · Claude Opus 4.5 leads on 6 of 31 shared benchmarks (Chatbot Arena Elo · Coding, Cl Bench, DeepResearch Bench, GSO-Bench, 2 more), Gemini 3.1 Pro Preview on 23 (APEX-Agents, ARC-AGI, ARC-AGI-2, Chatbot Arena Elo · Overall, 19 more) · 2 tied
- Price · Claude Opus 4.5 input $5.00/M vs Gemini 3.1 Pro Preview $2.00/M, Gemini 3.1 Pro Preview is 60% cheaper (OpenRouter list price, as of 2026-11-17)
- Cheapest provider · Claude Opus 4.5 at Amazon Bedrock and 4 other providers $5.00/M input (of 6 offers) as of 2026-11-17 · Gemini 3.1 Pro Preview at Google and 1 other provider $1.00/M input (of 6 offers) as of 2026-11-17
- Context · Claude Opus 4.5 200K tokens vs Gemini 3.1 Pro Preview 1.0M tokens
- Released · Claude Opus 4.5 Nov 24, 2025 vs Gemini 3.1 Pro Preview Feb 19, 2026 · 3 months apart
- Price history · Claude Opus 4.5 OpenRouter input list price had no move of 5% or more in the 90 days to 2026-11-17 · Gemini 3.1 Pro Preview OpenRouter input list price had no move of 5% or more in the 90 days to 2026-11-17
Sources · scores from public leaderboards, list prices and provider offers from OpenRouter, price moves from the BenchGecko changelog (5% or more, held in the next daily check), Gecko tests run by BenchGecko.
Hype vs Reality
Attention vs performance
Claude Opus 4.5
#199 by perf·#8 by attention
Gemini 3.1 Pro Preview
#143 by perf·#9 by attention
Best value
Gemini 3.1 Pro Preview
2.6x better value than Claude Opus 4.5
Claude Opus 4.5
2.7 pts/$
$15.00/M
Gemini 3.1 Pro Preview
7.0 pts/$
$7.00/M
Vendor risk
Who is behind the model
Anthropic
$1.40T·Tier 1
Google DeepMind
$3.49T·Tier 1
Head to head
31 benchmarks · 2 models
Claude Opus 4.5Gemini 3.1 Pro Preview
APEX-Agents
Gemini 3.1 Pro Preview leads by +16.7
APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments.
Claude Opus 4.5
18.4
Gemini 3.1 Pro Preview
35.1
ARC-AGI
Gemini 3.1 Pro Preview leads by +18.0
ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.
Claude Opus 4.5
80.0
Gemini 3.1 Pro Preview
98.0
ARC-AGI-2
Gemini 3.1 Pro Preview leads by +39.5
ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data.
Claude Opus 4.5
37.6
Gemini 3.1 Pro Preview
77.1
Chatbot Arena Elo · Coding
Claude Opus 4.5 leads by +22.3
Claude Opus 4.5
1468.5
Gemini 3.1 Pro Preview
1446.2
Chatbot Arena Elo · Overall
Gemini 3.1 Pro Preview leads by +17.2
Claude Opus 4.5
1469.8
Gemini 3.1 Pro Preview
1487.0
Balrog
Gemini 3.1 Pro Preview leads by +13.5
Balrog · benchmarks AI agents on text-based adventure games, testing language understanding, strategic planning, and long-horizon reasoning.
Claude Opus 4.5
43.5
Gemini 3.1 Pro Preview
57.0
Chess Puzzles
Gemini 3.1 Pro Preview leads by +36.4
Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities.
Claude Opus 4.5
12.0
Gemini 3.1 Pro Preview
48.4
Cl Bench
Claude Opus 4.5 leads by +0.3
Claude Opus 4.5
21.1
Gemini 3.1 Pro Preview
20.8
DeepResearch Bench
Claude Opus 4.5 leads by +7.0
DeepResearch Bench · evaluates AI on complex multi-step research tasks requiring information gathering, synthesis, and producing comprehensive analyses.
Claude Opus 4.5
54.8
Gemini 3.1 Pro Preview
47.8
Dtbench
Gemini 3.1 Pro Preview leads by +12.0
Claude Opus 4.5
83.1
Gemini 3.1 Pro Preview
95.1
Ebr Bench
Claude Opus 4.5
14.3
Gemini 3.1 Pro Preview
14.3
FrontierMath-2025-02-28-Private
Gemini 3.1 Pro Preview leads by +16.2
FrontierMath (Feb 2025) · original research-level math problems created by mathematicians, testing capabilities at the boundary of current AI mathematical reasoning.
Claude Opus 4.5
20.7
Gemini 3.1 Pro Preview
36.9
FrontierMath-Tier-4-2025-07-01-Private
Gemini 3.1 Pro Preview leads by +12.5
FrontierMath Tier 4 (Jul 2025) · the most challenging tier of frontier mathematics, containing problems that push the absolute limits of AI mathematical reasoning.
Claude Opus 4.5
4.2
Gemini 3.1 Pro Preview
16.7
FrontierMath-Tier-4-v2-Private
Gemini 3.1 Pro Preview leads by +21.9
Claude Opus 4.5
4.9
Gemini 3.1 Pro Preview
26.8
FrontierMath-Tiers-1-3-v2-Private
Gemini 3.1 Pro Preview leads by +26.6
Claude Opus 4.5
36.2
Gemini 3.1 Pro Preview
62.8
Furniture Assembly
Claude Opus 4.5
0.0
Gemini 3.1 Pro Preview
0.0
GeoBench
Gemini 3.1 Pro Preview leads by +9.0
GeoBench · tests geographic knowledge and spatial reasoning across countries, landmarks, coordinates, and geopolitical understanding.
Claude Opus 4.5
75.0
Gemini 3.1 Pro Preview
84.0
GPQA diamond
Gemini 3.1 Pro Preview leads by +15.0
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
Claude Opus 4.5
81.4
Gemini 3.1 Pro Preview
96.5
GSO-Bench
Claude Opus 4.5 leads by +3.9
GSO-Bench · evaluates AI models on real-world open-source software engineering tasks, testing the ability to understand and resolve actual GitHub issues.
Claude Opus 4.5
26.5
Gemini 3.1 Pro Preview
22.6
HLE
Gemini 3.1 Pro Preview leads by +22.3
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 4.5
21.4
Gemini 3.1 Pro Preview
43.7
Lmca
Gemini 3.1 Pro Preview leads by +11.0
Claude Opus 4.5
52.3
Gemini 3.1 Pro Preview
63.3
Metr Time Horizons
Gemini 3.1 Pro Preview leads by +2.1
Claude Opus 4.5
75.0
Gemini 3.1 Pro Preview
77.0
Mystery Game Puzzles
Gemini 3.1 Pro Preview leads by +13.2
Claude Opus 4.5
14.1
Gemini 3.1 Pro Preview
27.3
OTIS Mock AIME 2024-2025
Gemini 3.1 Pro Preview leads by +9.5
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Claude Opus 4.5
86.1
Gemini 3.1 Pro Preview
95.6
PostTrainBench
Gemini 3.1 Pro Preview leads by +0.8
Claude Opus 4.5
17.3
Gemini 3.1 Pro Preview
18.1
Proofbench
Claude Opus 4.5 leads by +10.0
Claude Opus 4.5
36.0
Gemini 3.1 Pro Preview
26.0
SimpleBench
Gemini 3.1 Pro Preview leads by +21.1
SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking.
Claude Opus 4.5
54.4
Gemini 3.1 Pro Preview
75.5
SimpleQA Verified
Gemini 3.1 Pro Preview leads by +31.7
SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information.
Claude Opus 4.5
41.8
Gemini 3.1 Pro Preview
73.5
SWE-Bench verified
Claude Opus 4.5 leads by +1.0
SWE-bench Verified · 500 human-validated tasks from 12 real Python repositories (Django, Flask, scikit-learn, sympy, and others). Each task requires the model to produce a git patch that resolves a real GitHub issue and passes the test suite. The verified subset eliminates ambiguous tasks from the original SWE-bench. The benchmark remains the most-cited evaluation for code-generation capability.
Claude Opus 4.5
76.7
Gemini 3.1 Pro Preview
75.6
Terminal Bench
Gemini 3.1 Pro Preview leads by +17.1
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 4.5
63.1
Gemini 3.1 Pro Preview
80.2
WeirdML
Gemini 3.1 Pro Preview leads by +8.3
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
Claude Opus 4.5
63.7
Gemini 3.1 Pro Preview
72.1
Full benchmark table
| Benchmark | Claude Opus 4.5 | Gemini 3.1 Pro Preview |
|---|---|---|
APEX-Agents APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments. | 18.4 | 35.1 |
ARC-AGI ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization. | 80.0 | 98.0 |
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. | 37.6 | 77.1 |
Chatbot Arena Elo · Coding | 1468.5 | 1446.2 |
Chatbot Arena Elo · Overall | 1469.8 | 1487.0 |
Balrog Balrog · benchmarks AI agents on text-based adventure games, testing language understanding, strategic planning, and long-horizon reasoning. | 43.5 | 57.0 |
Chess Puzzles Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities. | 12.0 | 48.4 |
Cl Bench | 21.1 | 20.8 |
DeepResearch Bench DeepResearch Bench · evaluates AI on complex multi-step research tasks requiring information gathering, synthesis, and producing comprehensive analyses. | 54.8 | 47.8 |
Dtbench | 83.1 | 95.1 |
Ebr Bench | 14.3 | 14.3 |
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. | 20.7 | 36.9 |
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. | 4.2 | 16.7 |
FrontierMath-Tier-4-v2-Private | 4.9 | 26.8 |
FrontierMath-Tiers-1-3-v2-Private | 36.2 | 62.8 |
Furniture Assembly | 0.0 | 0.0 |
GeoBench GeoBench · tests geographic knowledge and spatial reasoning across countries, landmarks, coordinates, and geopolitical understanding. | 75.0 | 84.0 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 81.4 | 96.5 |
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. | 26.5 | 22.6 |
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. | 21.4 | 43.7 |
Lmca | 52.3 | 63.3 |
Metr Time Horizons | 75.0 | 77.0 |
Mystery Game Puzzles | 14.1 | 27.3 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 86.1 | 95.6 |
PostTrainBench | 17.3 | 18.1 |
Proofbench | 36.0 | 26.0 |
SimpleBench SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking. | 54.4 | 75.5 |
SimpleQA Verified SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information. | 41.8 | 73.5 |
SWE-Bench verified SWE-bench Verified · 500 human-validated tasks from 12 real Python repositories (Django, Flask, scikit-learn, sympy, and others). Each task requires the model to produce a git patch that resolves a real GitHub issue and passes the test suite. The verified subset eliminates ambiguous tasks from the original SWE-bench. The benchmark remains the most-cited evaluation for code-generation capability. | 76.7 | 75.6 |
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. | 63.1 | 80.2 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 63.7 | 72.1 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| $5.00 | $25.00 | 200K tokens (~100 books) | $100.00 | |
| $2.00 | $12.00 | 1.0M tokens (~524 books) | $45.00 |
Claude Opus 4.5 vs Gemini 3.1 Pro Preview · questions answered from the data
Claude Opus 4.5 leads on 6 of 31 shared benchmarks (Chatbot Arena Elo · Coding, Cl Bench, DeepResearch Bench, GSO-Bench, 2 more), Gemini 3.1 Pro Preview on 23 (APEX-Agents, ARC-AGI, ARC-AGI-2, Chatbot Arena Elo · Overall, 19 more) · 2 tied