Head-to-head comparison

Claude Opus 4.5 vs. Gemini 3.1 Pro Preview

Compare published benchmark results from matching versions and cohorts, API pricing, and technical specifications. Any documented test setup differences remain visible.

shared result series: 19
distinct benchmarks: 7
latest retrieval: 2026-08-26

First modelSecond model

Specifications

General specifications

Release date, availability, context window, architecture, and published parameter counts at a glance. Undisclosed details remain marked as unavailable.

Specifications for Claude Opus 4.5 and Gemini 3.1 Pro Preview
Attribute	Claude Opus 4.5	Gemini 3.1 Pro Preview
Provider	Anthropic	Google
Released	Nov 2025	Feb 19, 2026
Availability	Active	Preview
Model type	Proprietary	Proprietary
Parameters	Not published	Not published
Architecture	Not published	Not published
Context window	200,000 tokens	1,048,576 tokens
Knowledge cutoff	March 2025	Not published
Technical sources	Model, Context	Model

Pricing

API pricing compared

The table compares official standard API rates per 1M tokens, including published cache pricing. Batch, fast, priority, regional surcharges, tool calls, and cloud platform pricing are excluded.

API pricing for Claude Opus 4.5 and Gemini 3.1 Pro Preview
Price type	Claude Opus 4.5	Gemini 3.1 Pro Preview
API input	$5	$2 ≤200K / $4 >200K
API output	$25	$12 ≤200K / $18 >200K
Cache read	$0.5	$0.2 ≤200K / $0.4 >200K
Cache write (5 min.)	$6.25	Not listed
Cache write (1 hr.)	$10	Not listed
Price verified	07/30/2026 Source	07/30/2026 Source

Claude Opus 4.5Gemini 3.1 Pro Preview

API inputUSD per 1M tokens, base tier

Claude Opus 4.5$5

Gemini 3.1 Pro Preview$2

API outputUSD per 1M tokens, base tier

Claude Opus 4.5$25

Gemini 3.1 Pro Preview$12

Cache readUSD per 1M tokens, base tier

Claude Opus 4.5$0.5

Gemini 3.1 Pro Preview$0.2

Key to the marks:better valuebetter, but not conclusivetie

Pricing by provider

Direct developer pricing and provider endpoint tariffs routed through OpenRouter are listed separately. Prices are in USD per 1M tokens. Context-dependent direct tiers plus regional, Flex, Priority, and other OpenRouter tariffs remain individually identifiable.

Provider pricing for Claude Opus 4.5 and Gemini 3.1 Pro Preview
Model	Provider	Purchase route and tariff	Input per 1M tokens	Output per 1M tokens	Source
Claude Opus 4.5	Anthropic	Direct from developerStandard API	Lowest listed price: $5	Lowest listed price: $25	Source07/30/2026
	Amazon Bedrock	Via OpenRouteramazon-bedrock	Lowest listed price: $5	Lowest listed price: $25	Source09/12/2026
	Amazon Bedrock	Via OpenRouteramazon-bedrock/eu-west-1	$5.5	$27.5	Source09/12/2026
	Anthropic	Via OpenRouteranthropic	Lowest listed price: $5	Lowest listed price: $25	Source09/12/2026
	Azure	Via OpenRouterazure/global	Lowest listed price: $5	Lowest listed price: $25	Source09/12/2026
	Claude Platform on AWS	Via OpenRouterclaude-on-aws	Lowest listed price: $5	Lowest listed price: $25	Source09/12/2026
	Google	Via OpenRoutergoogle-vertex/global	Lowest listed price: $5	Lowest listed price: $25	Source09/12/2026
Gemini 3.1 Pro Preview	Google	Direct from developerUp to 200,000 context tokens	$2	$12	Source07/30/2026
	Google	Direct from developerAbove 200,000 context tokens	$4	$18	Source07/30/2026
	Google	Via OpenRoutergoogle-vertex/global	$2	$12	Source09/12/2026
	Google	Via OpenRoutergoogle-vertex/global/flex	Lowest listed price: $1	Lowest listed price: $6	Source09/12/2026
	Google	Via OpenRoutergoogle-vertex/global/priority	$3.6	$21.6	Source09/12/2026
	Google AI Studio	Via OpenRoutergoogle-ai-studio	$2	$12	Source09/12/2026
	Google AI Studio	Via OpenRoutergoogle-ai-studio/flex	Lowest listed price: $1	Lowest listed price: $6	Source09/12/2026
	Google AI Studio	Via OpenRoutergoogle-ai-studio/priority	$3.6	$21.6	Source09/12/2026

Performance

Shared benchmarks

Only published results with the same benchmark version, task, metric, and comparison cohort are paired. Different reasoning levels, agents, harnesses, or output limits appear directly in the table.

Capability profile from matched benchmarks

Each axis averages directly matched benchmark families. A value of 100 means the stronger value within this pair, not a universal quality score.

Not enough directly matched categories for a reliable radar

Only 1 of at least 5 categories have comparable, direct model measurements. The benchmark tables below show the available evidence without filling missing axes.

Claude Opus 4.5Gemini 3.1 Pro Preview

GAIA ReliabilityAccuracy. Higher is better. No documented setup difference

Claude Opus 4.568.5 %

Gemini 3.1 Pro Preview76.2 %

MultiChallengeOverall. Higher is better. Test setup not documented as identical

Claude Opus 4.558.97% ± 0.42

Gemini 3.1 Pro Preview71.37% ± 1.74

MultiNRCOverall. Higher is better. Test setup not documented as identical

Claude Opus 4.541.23% ± 2.97

Gemini 3.1 Pro Preview64.74% ± 2.88

PRBench FinanceOverall. Higher is better. Test setup not documented as identical

Claude Opus 4.546.16% ± 0.27

Gemini 3.1 Pro Preview41.87% ± 1.22

PRBench LegalOverall. Higher is better. Test setup not documented as identical

Claude Opus 4.544.21% ± 0.34

Gemini 3.1 Pro Preview44.02% ± 1.12

TutorBenchOverall. Higher is better. Test setup not documented as identical

Claude Opus 4.549.82% ± 0.98

Gemini 3.1 Pro Preview52.99% ± 1.06

τ-bench Airline CleanAccuracy. Higher is better. No documented setup difference

Claude Opus 4.580.8 %

Gemini 3.1 Pro Preview82.1 %

GAIA ReliabilityConsistency. Higher is better. No documented setup difference

Claude Opus 4.50.81 rating points

Gemini 3.1 Pro Preview0.76 rating points

MultiNRCOverall. Higher is better. Test setup not documented as identical

Claude Opus 4.548.63% ± 3.02

Gemini 3.1 Pro Preview64.74% ± 2.88

TutorBenchOverall. Higher is better. Test setup not documented as identical

Claude Opus 4.551.2% ± 0.99

Gemini 3.1 Pro Preview52.99% ± 1.06

τ-bench Airline CleanConsistency. Higher is better. No documented setup difference

Claude Opus 4.50.84 rating points

Gemini 3.1 Pro Preview0.81 rating points

GAIA ReliabilityPredictability. Higher is better. No documented setup difference

Claude Opus 4.50.84 rating points

Gemini 3.1 Pro Preview0.78 rating points

Key to the marks:better valuebetter, but not conclusivetie

Benchmark scores for Claude Opus 4.5 and Gemini 3.1 Pro Preview
Benchmark and task	Claude Opus 4.5	Gemini 3.1 Pro Preview	Source
GAIA ReliabilityAccuracy Test details No documented difference in the test setupVersion: final leak-cleaned public validation setMetric: exact-match accuracyScoring: Higher is betterSettings: Agent: HAL Reliability, Harness: HAL Reliability165 GAIA tasks with repeated runs and perturbations. HAL does not publish cost for this evaluation.	68.5 %	Winner76.2 %	Princeton HAL Source details Data as of: 08/06/2026
MultiChallengeOverall Test details Test setup not documented as identicalNo direct winnerVersion: four multi-turn challenge typesMetric: rubric pass rateScoring: Higher is betterClaude Opus 4.5: Reasoning: thinking, Attempts: pass@1 Gemini 3.1 Pro Preview: Attempts: pass@1Multi-turn conversations with binary task-specific rubrics.	58.97% ± 0.42 95% confidence interval: ±0.42	Higher measured value71.37% ± 1.74 95% confidence interval: ±1.74	Scale AI Source details Data as of: 08/26/2026
MultiNRCOverall Test details Test setup not documented as identicalNo direct winnerVersion: French, Spanish, and Chinese releaseMetric: accuracyScoring: Higher is betterClaude Opus 4.5: Reasoning: none, Attempts: pass@1 Gemini 3.1 Pro Preview: Attempts: pass@1Native, non-translated reasoning tasks with automatic evaluation.	41.23% ± 2.97 95% confidence interval: ±2.97	Higher measured value64.74% ± 2.88 95% confidence interval: ±2.88	Scale AI Source details Data as of: 08/26/2026
PRBench FinanceOverall Test details Test setup not documented as identicalNo direct winnerVersion: full finance setMetric: weighted rubric scoreScoring: Higher is betterClaude Opus 4.5: Reasoning: thinking, Attempts: pass@1 Gemini 3.1 Pro Preview: Attempts: pass@1600 finance tasks with task-specific rubrics and o4-mini as judge.	Higher measured value46.16% ± 0.27 95% confidence interval: ±0.27	41.87% ± 1.22 95% confidence interval: ±1.22	Scale AI Source details Data as of: 08/26/2026
PRBench LegalOverall Test details Test setup not documented as identicalNo direct winnerVersion: full legal setMetric: weighted rubric scoreScoring: Higher is betterClaude Opus 4.5: Reasoning: thinking, Attempts: pass@1 Gemini 3.1 Pro Preview: Attempts: pass@1500 legal tasks with task-specific rubrics and o4-mini as judge.	Higher measured value44.21% ± 0.34 95% confidence interval: ±0.34	44.02% ± 1.12 95% confidence interval: ±1.12	Scale AI Source details Data as of: 08/26/2026
TutorBenchOverall Test details Test setup not documented as identicalNo direct winnerVersion: 1,490-task releaseMetric: weighted rubric scoreScoring: Higher is betterClaude Opus 4.5: Reasoning: none, Attempts: pass@1 Gemini 3.1 Pro Preview: Attempts: pass@11,490 tasks with weighted rubrics and Claude 4 Sonnet as judge.	49.82% ± 0.98 95% confidence interval: ±0.98	Higher measured value52.99% ± 1.06 95% confidence interval: ±1.06	Scale AI Source details Data as of: 08/26/2026
τ-bench Airline CleanAccuracy Test details No documented difference in the test setupVersion: 26-task cleaned subsetMetric: task success rateScoring: Higher is betterSettings: Agent: HAL Reliability, Harness: HAL Reliability26 cleaned airline tasks. Princeton HAL recomputed every metric from scratch on this subset.	80.8 %	Winner82.1 %	Princeton HAL Source details Data as of: 08/05/2026
GAIA ReliabilityConsistency Test details No documented difference in the test setupVersion: final leak-cleaned public validation setMetric: aggregate consistencyScoring: Higher is betterSettings: Agent: HAL Reliability, Harness: HAL Reliability165 GAIA tasks with repeated runs and perturbations. HAL does not publish cost for this evaluation.	Winner0.81 rating points	0.76 rating points	Princeton HAL Source details Data as of: 08/06/2026
MultiNRCOverall Test details Test setup not documented as identicalNo direct winnerVersion: French, Spanish, and Chinese releaseMetric: accuracyScoring: Higher is betterClaude Opus 4.5: Reasoning: thinking, Attempts: pass@1 Gemini 3.1 Pro Preview: Attempts: pass@1Native, non-translated reasoning tasks with automatic evaluation.	48.63% ± 3.02 95% confidence interval: ±3.02	Higher measured value64.74% ± 2.88 95% confidence interval: ±2.88	Scale AI Source details Data as of: 08/26/2026
TutorBenchOverall Test details Test setup not documented as identicalNo direct winnerVersion: 1,490-task releaseMetric: weighted rubric scoreScoring: Higher is betterClaude Opus 4.5: Reasoning: thinking, Attempts: pass@1 Gemini 3.1 Pro Preview: Attempts: pass@11,490 tasks with weighted rubrics and Claude 4 Sonnet as judge.	51.2% ± 0.99 95% confidence interval: ±0.99	Higher measured value52.99% ± 1.06 95% confidence interval: ±1.06	Scale AI Source details Data as of: 08/26/2026
τ-bench Airline CleanConsistency Test details No documented difference in the test setupVersion: 26-task cleaned subsetMetric: aggregate consistencyScoring: Higher is betterSettings: Agent: HAL Reliability, Harness: HAL Reliability26 cleaned airline tasks. Princeton HAL recomputed every metric from scratch on this subset.	Winner0.84 rating points	0.81 rating points	Princeton HAL Source details Data as of: 08/05/2026
GAIA ReliabilityPredictability Test details No documented difference in the test setupVersion: final leak-cleaned public validation setMetric: aggregate predictabilityScoring: Higher is betterSettings: Agent: HAL Reliability, Harness: HAL Reliability165 GAIA tasks with repeated runs and perturbations. HAL does not publish cost for this evaluation.	Winner0.84 rating points	0.78 rating points	Princeton HAL Source details Data as of: 08/06/2026
τ-bench Airline CleanPredictability Test details No documented difference in the test setupVersion: 26-task cleaned subsetMetric: aggregate predictabilityScoring: Higher is betterSettings: Agent: HAL Reliability, Harness: HAL Reliability26 cleaned airline tasks. Princeton HAL recomputed every metric from scratch on this subset.	Winner0.85 rating points	0.83 rating points	Princeton HAL Source details Data as of: 08/05/2026
GAIA ReliabilityReliability Test details No documented difference in the test setupVersion: final leak-cleaned public validation setMetric: aggregate reliabilityScoring: Higher is betterSettings: Agent: HAL Reliability, Harness: HAL Reliability165 GAIA tasks with repeated runs and perturbations. HAL does not publish cost for this evaluation.	Winner0.85 rating points	0.82 rating points	Princeton HAL Source details Data as of: 08/06/2026
τ-bench Airline CleanReliability Test details No documented difference in the test setupVersion: 26-task cleaned subsetMetric: aggregate reliabilityScoring: Higher is betterSettings: Agent: HAL Reliability, Harness: HAL Reliability26 cleaned airline tasks. Princeton HAL recomputed every metric from scratch on this subset.	Winner0.88 rating points	0.86 rating points	Princeton HAL Source details Data as of: 08/05/2026
GAIA ReliabilityRobustness Test details No documented difference in the test setupVersion: final leak-cleaned public validation setMetric: aggregate robustnessScoring: Higher is betterSettings: Agent: HAL Reliability, Harness: HAL Reliability165 GAIA tasks with repeated runs and perturbations. HAL does not publish cost for this evaluation.	0.91 rating points	Winner0.94 rating points	Princeton HAL Source details Data as of: 08/06/2026
τ-bench Airline CleanRobustness Test details No documented difference in the test setupVersion: 26-task cleaned subsetMetric: aggregate robustnessScoring: Higher is betterSettings: Agent: HAL Reliability, Harness: HAL Reliability26 cleaned airline tasks. Princeton HAL recomputed every metric from scratch on this subset.	Winner0.97 rating points	0.94 rating points	Princeton HAL Source details Data as of: 08/05/2026
GAIA ReliabilitySafety Test details No documented difference in the test setupVersion: final leak-cleaned public validation setMetric: aggregate safetyScoring: Higher is betterSettings: Agent: HAL Reliability, Harness: HAL Reliability165 GAIA tasks with repeated runs and perturbations. HAL does not publish cost for this evaluation.	Tie1 rating points	Tie1 rating points	Princeton HAL Source details Data as of: 08/06/2026
τ-bench Airline CleanSafety Test details No documented difference in the test setupVersion: 26-task cleaned subsetMetric: aggregate safetyScoring: Higher is betterSettings: Agent: HAL Reliability, Harness: HAL Reliability26 cleaned airline tasks. Princeton HAL recomputed every metric from scratch on this subset.	Winner0.98 rating points	0.97 rating points	Princeton HAL Source details Data as of: 08/05/2026
Key to the marks:better valuebetter, but not conclusivetie

Our model tests

Community tests you can inspect

Compare how the models respond to the same prompt. Each test shows the first attempt, with no subsequent fixes to the generated code. These results do not contribute to an overall score.

Pelican on a bicycle

A community classic for free-form SVG drawing.

Claude Opus 4.5

Not tested yet

Gemini 3.1 Pro Preview

API access was unavailable for this run. The model was not evaluated.

First attempt

Run date: 7 Sept 2026
API provider: openrouter
Requested API model: google/gemini-3.1-pro-preview
Pinned endpoint: Google AI Studio | google/gemini-3.1-pro-preview-20260219
Reasoning setting: high
Temperature: Provider default
Token limit including reasoning: 8,192 tokens
Test protocol: community-visual-tests-v1
Technical checks: Not run

Not yet visually reviewed. A successful recording does not confirm correct physics or full compliance with the prompt.

Prompt and test conditions

Generate an SVG of a pelican riding a bicycle

Token limit including reasoning: 8,192 tokens

Task origin (Simon Willison)

How to read this comparison

Winning one benchmark is not an overall verdict

Your workload, budget, and required context length matter most. A coding benchmark says little about visual understanding or agent performance.

The standard price table uses direct developer rates. The provider comparison labels OpenRouter endpoints, regions, and special tariffs separately.

Every result links to its measurement source and retrieval date. Labels distinguish vendor reports, official benchmark leaderboards, and independent evaluations. Disputed or archived results are excluded.

More matchups

Compare Claude Opus 4.5 and Gemini 3.1 Pro Preview with other leading models using the same data and benchmark logic.

View the complete LLM comparison