Skip to main content

Head-to-head comparison

Gemini 3.5 Flash 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
198
distinct benchmarks
31
latest retrieval
2026-09-04

Change model selection

Specifications

General specifications

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

Specifications for Gemini 3.5 Flash and Gemini 3.1 Pro Preview
AttributeGemini 3.5 FlashGemini 3.1 Pro Preview
ProviderGoogleGoogle
ReleasedMay 19, 2026Feb 19, 2026
AvailabilityActivePreview
Model typeProprietaryProprietary
ParametersNot publishedNot published
ArchitectureNot publishedNot published
Context window1,048,576 tokens1,048,576 tokens
Knowledge cutoffNot publishedNot published
Technical sourcesModelModel

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 Gemini 3.5 Flash and Gemini 3.1 Pro Preview
Price typeGemini 3.5 FlashGemini 3.1 Pro Preview
API input$1.5$2 ≤200K / $4 >200K
API output$9$12 ≤200K / $18 >200K
Cache read$0.15$0.2 ≤200K / $0.4 >200K
Cache write (5 min.)Not listedNot listed
Cache write (1 hr.)Not listedNot listed
Price verified07/30/2026 Source07/30/2026 Source
Gemini 3.5 FlashGemini 3.1 Pro Preview
API inputUSD per 1M tokens, base tier
Gemini 3.5 FlashLower price$1.5
Gemini 3.1 Pro Preview$2
API outputUSD per 1M tokens, base tier
Gemini 3.5 FlashLower price$9
Gemini 3.1 Pro Preview$12
Cache readUSD per 1M tokens, base tier
Gemini 3.5 FlashLower price$0.15
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 Gemini 3.5 Flash and Gemini 3.1 Pro Preview
ModelProviderPurchase route and tariffInput per 1M tokensOutput per 1M tokensSource
Gemini 3.5 FlashGoogleDirect from developerStandard API$1.5$9Source07/30/2026
GoogleVia OpenRoutergoogle-vertex/global$1.5$9Source09/08/2026
GoogleVia OpenRoutergoogle-vertex/global/flexLowest listed price: $0.75Lowest listed price: $4.5Source09/08/2026
GoogleVia OpenRoutergoogle-vertex/global/priority$2.7$16.2Source09/08/2026
GoogleVia OpenRoutergoogle-vertex/us$1.65$9.9Source09/08/2026
Google AI StudioVia OpenRoutergoogle-ai-studio$1.5$9Source09/08/2026
Google AI StudioVia OpenRoutergoogle-ai-studio/flexLowest listed price: $0.75Lowest listed price: $4.5Source09/08/2026
Google AI StudioVia OpenRoutergoogle-ai-studio/priority$2.7$16.2Source09/08/2026
Gemini 3.1 Pro PreviewGoogleDirect from developerUp to 200,000 context tokens$2$12Source07/30/2026
GoogleDirect from developerAbove 200,000 context tokens$4$18Source07/30/2026
GoogleVia OpenRoutergoogle-vertex/global$2$12Source09/08/2026
GoogleVia OpenRoutergoogle-vertex/global/flexLowest listed price: $1Lowest listed price: $6Source09/08/2026
GoogleVia OpenRoutergoogle-vertex/global/priority$3.6$21.6Source09/08/2026
Google AI StudioVia OpenRoutergoogle-ai-studio$2$12Source09/08/2026
Google AI StudioVia OpenRoutergoogle-ai-studio/flexLowest listed price: $1Lowest listed price: $6Source09/08/2026
Google AI StudioVia OpenRoutergoogle-ai-studio/priority$3.6$21.6Source09/08/2026

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.

255075100CodingReasoningMathematicsKnowledgeFinanceLegalAgentic tasksMultimodal
Gemini 3.5 Flash
Gemini 3.1 Pro Preview
Coding: 99.6 / 83.5 (5 benchmark families)
Reasoning: 83.1 / 100 (2 benchmark families)
Mathematics: 100 / 83.9 (1 benchmark family)
Knowledge: 94.6 / 100 (2 benchmark families)
Finance: 99.4 / 90.2 (5 benchmark families)
Legal: 98.3 / 59.4 (3 benchmark families)
Agentic tasks: 99.6 / 98.3 (3 benchmark families)
Multimodal: 100 / 99.9 (1 benchmark family)

122 matched rows from 29 benchmark families were considered. The radar shows 8 of 13 comparable categories.

Gemini 3.5 FlashGemini 3.1 Pro Preview
Code MigrationOverall. Higher is better. No documented setup difference
Gemini 3.5 FlashWinner26.745 %
Gemini 3.1 Pro Preview17.313 %
CorpFin v2Overall. Higher is better. No documented setup difference
Gemini 3.5 FlashWinner64.686 %
Gemini 3.1 Pro Preview64.491 %
CyberBenchOverall. Higher is better. No documented setup difference
Gemini 3.5 FlashWinner69.331 %
Gemini 3.1 Pro Preview36.441 %
EMBOverall. Higher is better. No documented setup difference
Gemini 3.5 FlashWinner63.548 %
Gemini 3.1 Pro Preview52.616 %
EnigmaEvalOverall. Higher is better. No documented setup difference
Gemini 3.5 Flash25.41% ± 2.45
Gemini 3.1 Pro PreviewWinner36.78% ± 2.71
Finance Agent (v2)Overall. Higher is better. No documented setup difference
Gemini 3.5 FlashWinner57.861 %
Gemini 3.1 Pro Preview42.982 %
GAIA ReliabilityAccuracy. Higher is better. No documented setup difference
Gemini 3.5 FlashWinner79.2 %
Gemini 3.1 Pro Preview76.2 %
GPQA DiamondOverall. Higher is better. No documented setup difference
Gemini 3.5 Flash92.676 %
Gemini 3.1 Pro PreviewWinner95.454 %
Harvey's Legal Agent BenchmarkOverall · Task fully resolved. Higher is better. No documented setup difference
Gemini 3.5 FlashWinner2.5 %
Gemini 3.1 Pro Preview0 %
Legal Research BenchOverall · All-pass. Higher is better. No documented setup difference
Gemini 3.5 FlashWinner30.769 %
Gemini 3.1 Pro Preview20.673 %
LegalBenchOverall. Higher is better. No documented setup difference
Gemini 3.5 Flash83.602 %
Gemini 3.1 Pro PreviewWinner87.398 %
LiveCodeBench v6Overall. Higher is better. No documented setup difference
Gemini 3.5 Flash87.604 %
Gemini 3.1 Pro PreviewWinner88.485 %

Key to the marks:better valuebetter, but not conclusivetie

Benchmark scores for Gemini 3.5 Flash and Gemini 3.1 Pro Preview
Benchmark and taskGemini 3.5 FlashGemini 3.1 Pro PreviewSource
Code MigrationOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
Winner26.745 %
17.313 %
Vals AI
Source details
Data as of: 09/03/2026
CorpFin v2Overall
Test details
No documented difference in the test setupVersion: 2Metric: accuracyScoring: Higher is better
Winner64.686 %
64.491 %
Vals AI
Source details
Data as of: 08/12/2026
CyberBenchOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
Winner69.331 %
36.441 %
Vals AI
Source details
Data as of: 09/01/2026
EMBOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
Winner63.548 %
52.616 %
Vals AI
Source details
Data as of: 09/03/2026
EnigmaEvalOverall
Test details
No documented difference in the test setupVersion: open-source release 2026-07-23Metric: pass@1 accuracyScoring: Higher is betterGemini 3.5 Flash: Snapshot: gemini-3.5-flash-high, Reasoning: high, Attempts: pass@1
Gemini 3.1 Pro Preview: Snapshot: gemini-3.1-pro-preview-high, Reasoning: high, Attempts: pass@1
Complex multimodal puzzles with exact answer matching and one attempt.
25.41% ± 2.45
95% confidence interval: ±2.45
Winner36.78% ± 2.71
95% confidence interval: ±2.71
Scale AI
Source details
Data as of: 07/23/2026
Finance Agent (v2)Overall
Test details
No documented difference in the test setupVersion: 2Metric: accuracyScoring: Higher is better
Winner57.861 %
42.982 %
Vals AI
Source details
Data as of: 09/03/2026
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.
Winner79.2 %
76.2 %
Princeton HAL
Source details
Data as of: 08/06/2026
GPQA DiamondOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
92.676 %
Winner95.454 %
Vals AI
Source details
Data as of: 09/01/2026
Harvey's Legal Agent BenchmarkOverall · Task fully resolved
Test details
No documented difference in the test setupVersion: 1Metric: task resolution rateScoring: Higher is better
Winner2.5 %
0 %
Vals AI
Source details
Data as of: 09/03/2026
Legal Research BenchOverall · All-pass
Test details
No documented difference in the test setupVersion: 1Metric: all-pass rateScoring: Higher is better
Winner30.769 %
20.673 %
Vals AI
Source details
Data as of: 09/03/2026
LegalBenchOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
83.602 %
Winner87.398 %
Vals AI
Source details
Data as of: 09/01/2026
LiveCodeBench v6Overall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
87.604 %
Winner88.485 %
Vals AI
Source details
Data as of: 09/01/2026
LMArena Text, Style ControlOverall
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGemini 3.5 Flash: Snapshot: gemini-3.5-flash-high, Reasoning: high, Harness: Arena text, style control
Gemini 3.1 Pro Preview: Snapshot: gemini-3.1-pro-preview, Harness: Arena text, style control
Gemini 3.5 Flash: Style-controlled Arena rating from 33,874 anonymous pairwise votes.Gemini 3.1 Pro Preview: Style-controlled Arena rating from 102,763 anonymous pairwise votes.
1,479.148
95% confidence interval: ±4.345The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Higher measured value1,486.746
95% confidence interval: ±3.217The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
LMArena
Source details
Data as of: 09/01/2026
MedCodeOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
55.825 %
Winner59.062 %
Vals AI
Source details
Data as of: 09/03/2026
MedScribeOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
Winner76.574 %
76.114 %
Vals AI
Source details
Data as of: 09/03/2026
MMLU-ProOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
89.515 %
Winner90.987 %
Vals AI
Source details
Data as of: 09/01/2026
MMMU-ProOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
Winner88.266 %
88.208 %
Vals AI
Source details
Data as of: 09/01/2026
MortgageTaxOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
68.124 %
Winner69.396 %
Vals AI
Source details
Data as of: 09/01/2026
ProgramBenchOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
Tie0 %
Tie0 %
Vals AI
Source details
Data as of: 09/01/2026
ProofBenchOverall
Test details
No documented difference in the test setupVersion: 1.1Metric: accuracyScoring: Higher is better
Winner31 %
26 %
Vals AI
Source details
Data as of: 09/01/2026
Public Benefits Bench v1.1Overall
Test details
No documented difference in the test setupVersion: 1.1Metric: accuracyScoring: Higher is better
Winner59.472 %
53.789 %
Vals AI
Source details
Data as of: 09/01/2026
SAGEOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
Winner49.885 %
48.677 %
Vals AI
Source details
Data as of: 09/03/2026
SimpleQA Verified v2Kaggle score
Test details
No documented difference in the test setupVersion: Kaggle leaderboard 2026-08-03Metric: Kaggle leaderboard scoreScoring: Higher is better1,000 verified prompts without tools. Kaggle independently reproduced the results.
70.4 %
95% confidence interval: ±2.9
Winner77.5 %
95% confidence interval: ±2.6
Kaggle
Source details
Data as of: 08/03/2026
SWE-bench VerifiedOverall
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
Tie78.8 %
Tie78.8 %
Vals AI
Source details
Data as of: 09/01/2026
TaxEval v2Overall
Test details
No documented difference in the test setupVersion: 2Metric: accuracyScoring: Higher is better
Winner74.366 %
72.882 %
Vals AI
Source details
Data as of: 09/01/2026
Terminal-Bench 2.0Overall
Test details
No documented difference in the test setupVersion: 2.0Metric: accuracyScoring: Higher is better
Tie67.416 %
Tie67.416 %
Vals AI
Source details
Data as of: 06/04/2026
Terminal-Bench 2.1Overall
Test details
No documented difference in the test setupVersion: 2.1Metric: accuracyScoring: Higher is better
Winner74.157 %
70.787 %
Vals AI
Source details
Data as of: 09/03/2026
Vals IndexOverall
Test details
No documented difference in the test setupVersion: 2Metric: weighted index scoreScoring: Higher is better
Winner53.079 %
41.903 %
Vals AI
Source details
Data as of: 09/03/2026
Vals Multimodal IndexOverall
Test details
No documented difference in the test setupVersion: 1.2Metric: weighted index scoreScoring: Higher is better
Winner62.928 %
56.068 %
Vals AI
Source details
Data as of: 08/11/2026
Vibe Code Bench v1.1Overall
Test details
No documented difference in the test setupVersion: 1.1Metric: accuracyScoring: Higher is betterSettings: Harness: OpenHands
Winner48.683 %
32.034 %
Vals AI
Source details
Data as of: 09/03/2026
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
Code MigrationCLI
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Winner17.175 %
10.833 %
Vals AI
Source details
Data as of: 08/09/2026
CorpFin v2Exact Pages
Test details
No documented difference in the test setupVersion: 2Metric: accuracyScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Winner62.821 %
60.839 %
Vals AI
Source details
Data as of: 08/06/2026
CyberBenchPatch
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
Winner81.034 %
69.492 %
Vals AI
Source details
Data as of: 08/03/2026
EMBComps
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Winner69.788 %
44.505 %
Vals AI
Source details
Data as of: 08/06/2026
Finance Agent (v2)Adjustments
Test details
No documented difference in the test setupVersion: 2Metric: accuracyScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Winner54.869 %
38.361 %
Vals AI
Source details
Data as of: 08/06/2026
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.
0.75 rating points
Winner0.76 rating points
Princeton HAL
Source details
Data as of: 08/06/2026
GPQA DiamondFew-Shot CoT
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterGemini 3.5 Flash: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Gemini 3.1 Pro Preview: Reasoning: high, Temperature: 1, Output limit: 65,535 tokens
91.414 %
Higher measured value95.96 %
Vals AI
Source details
Data as of: 08/09/2026
Legal Research BenchAdministrative / Regulatory · All-pass
Test details
No documented difference in the test setupVersion: 1Metric: all-pass rateScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Winner43.939 %
24.242 %
Vals AI
Source details
Data as of: 08/06/2026
LegalBenchConclusion Tasks
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
81.939 %
Winner89.257 %
Vals AI
Source details
Data as of: 08/09/2026
LiveCodeBench v6Easy
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Tie99.068 %
Tie99.068 %
Vals AI
Source details
Data as of: 08/09/2026
LMArena Text, Style ControlOverall
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGemini 3.5 Flash: Snapshot: gemini-3.5-flash-medium, Reasoning: medium, Harness: Arena text, style control
Gemini 3.1 Pro Preview: Snapshot: gemini-3.1-pro-preview, Harness: Arena text, style control
Gemini 3.5 Flash: Style-controlled Arena rating from 32,274 anonymous pairwise votes.Gemini 3.1 Pro Preview: Style-controlled Arena rating from 102,763 anonymous pairwise votes.
1,474.438
95% confidence interval: ±4.676The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Higher measured value1,486.746
95% confidence interval: ±3.217The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
LMArena
Source details
Data as of: 09/01/2026
MMLU-ProBiology
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterGemini 3.5 Flash: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Gemini 3.1 Pro Preview: Reasoning: high, Temperature: 1, Output limit: 65,535 tokens
93.863 %
Higher measured value95.258 %
Vals AI
Source details
Data as of: 08/09/2026
MortgageTaxNumerical Extraction
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Winner77.504 %
76.391 %
Vals AI
Source details
Data as of: 08/09/2026
ProgramBenchAlmost Resolved
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Winner1.5 %
1 %
Vals AI
Source details
Data as of: 08/09/2026
SAGEAutomata
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 30,000 tokens
Winner38.492 %
32.143 %
Vals AI
Source details
Data as of: 08/09/2026
SWE-bench Verified<15 min fix
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
86.598 %
Winner88.66 %
Vals AI
Source details
Data as of: 08/08/2026
TaxEval v2Correctness
Test details
Test setup not documented as identicalNo direct winnerVersion: 2Metric: accuracyScoring: Higher is betterGemini 3.5 Flash: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Gemini 3.1 Pro Preview: Reasoning: high, Temperature: 1, Output limit: 65,535 tokens
Higher measured value61.897 %
60.749 %
Vals AI
Source details
Data as of: 08/09/2026
Terminal-Bench 2.0Easy
Test details
No documented difference in the test setupVersion: 2.0Metric: accuracyScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Tie100 %
Tie100 %
Vals AI
Source details
Data as of: 06/04/2026
Terminal-Bench 2.1Easy
Test details
No documented difference in the test setupVersion: 2.1Metric: accuracyScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Winner100 %
91.667 %
Vals AI
Source details
Data as of: 08/06/2026
Vals IndexCorpFin v2
Test details
No documented difference in the test setupVersion: 1.2Metric: benchmark scoreScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
64.569 %
Winner65.734 %
Vals AI
Source details
Data as of: 08/09/2026
Vals Multimodal IndexCorpFin v2
Test details
No documented difference in the test setupVersion: 1.2Metric: benchmark scoreScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
64.569 %
Winner65.734 %
Vals AI
Source details
Data as of: 08/09/2026
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.
0.78 rating points
Winner0.81 rating points
Princeton HAL
Source details
Data as of: 08/05/2026
Code MigrationCOBOL
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Winner55.455 %
36.753 %
Vals AI
Source details
Data as of: 08/09/2026
CorpFin v2Max Fitting Context
Test details
No documented difference in the test setupVersion: 2Metric: accuracyScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
66.667 %
Winner66.9 %
Vals AI
Source details
Data as of: 08/06/2026
CyberBenchPoC
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is better
Winner57.627 %
3.39 %
Vals AI
Source details
Data as of: 08/03/2026
EMBDataroom
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Winner80.921 %
79.17 %
Vals AI
Source details
Data as of: 08/06/2026
Finance Agent (v2)All-Pass
Test details
No documented difference in the test setupVersion: 2Metric: accuracyScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Winner45.706 %
30.879 %
Vals AI
Source details
Data as of: 08/06/2026
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.
Winner0.8 rating points
0.78 rating points
Princeton HAL
Source details
Data as of: 08/06/2026
GPQA DiamondZero-Shot CoT
Test details
Test setup not documented as identicalNo direct winnerVersion: 1Metric: accuracyScoring: Higher is betterGemini 3.5 Flash: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Gemini 3.1 Pro Preview: Reasoning: high, Temperature: 1, Output limit: 65,535 tokens
93.939 %
Higher measured value94.949 %
Vals AI
Source details
Data as of: 08/09/2026
Legal Research BenchAdministrative / Regulatory · Weighted partial credit
Test details
No documented difference in the test setupVersion: 1Metric: weighted partial-credit scoreScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Winner87.126 %
79.215 %
Vals AI
Source details
Data as of: 08/06/2026
LegalBenchInterpretation Tasks
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
Winner88.675 %
83.823 %
Vals AI
Source details
Data as of: 08/09/2026
LiveCodeBench v6Hard
Test details
No documented difference in the test setupVersion: 1Metric: accuracyScoring: Higher is betterSettings: Reasoning: high, Temperature: 1, Output limit: 65,536 tokens
73.143 %
Winner76.571 %
Vals AI
Source details
Data as of: 08/09/2026
LMArena Text, Style ControlBusiness, management and finance
Test details
Test setup not documented as identicalNo direct winnerVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGemini 3.5 Flash: Snapshot: gemini-3.5-flash-high, Reasoning: high, Harness: Arena text, style control
Gemini 3.1 Pro Preview: Snapshot: gemini-3.1-pro-preview, Harness: Arena text, style control
Gemini 3.5 Flash: Style-controlled Arena rating from 6,750 anonymous pairwise votes.Gemini 3.1 Pro Preview: Style-controlled Arena rating from 20,283 anonymous pairwise votes.
1,465.666
95% confidence interval: ±7.938The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
Higher measured value1,476.52
95% confidence interval: ±5.472The Arena score measures human preference rather than a fixed capability test. The confidence interval and model configuration are part of the result.
LMArena
Source details
Data as of: 09/01/2026

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.

Gemini 3.5 Flash

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-3.5-flash
Pinned endpoint
Google AI Studio | google/gemini-3.5-flash-20260519
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.

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 Gemini 3.5 Flash and Gemini 3.1 Pro Preview with other leading models using the same data and benchmark logic.

View the complete LLM comparison