AIEvals

Models

Gemini 3.1 Pro

Google’s Gemini 3.1 Pro preview, with published multimodal, scientific and coding results.

GoogleReleased 19 February 2026Context 1M tokensOutput 66k tokensPrivate weights$2.00 in / $12.00 out per Mtok

Family Gemini 3.148 benchmarks with a published result1 independent first place among the 46 models this site listsRead 17 October 2026

Where it stands

Rank compares tracked models with a score. Published shows the source’s rank and field size, when available.

BenchmarkScoreRankPublishedLeaderBehind
Vals IndexVals AI33.44%32of 3337of 41Gemini 4 Argon35.46 pts
Artificial Analysis Intelligence IndexArtificial Analysis3039of 42no list readClaude Opus 5.528 pts
Terminal-Bench 4.0Artificial Analysis4.04%37of 40no list readClaude Sonnet 5.559.60 pts
SWE-bench (Verified)Vals AI78.80%20of 2328of 88Claude Opus 518.20 pts
LiveCodeBenchVals AI88.49%6of 216of 143Claude Fable 5.12.03 pts
GPQA DiamondVals AI95.45%1of 231of 138Gemini 3.1 Pro—
MMLU ProVals AI90.99%4of 244of 138Claude Fable 5.11.39 pts
MMMU ProVals AI88.21%8of 2010of 93Claude Fable 5.12.43 pts
Vibe Code Bench v1.1Vals AI32.03%35of 3863of 106Claude Sonnet 5.560.36 pts
ProgramBenchVals AI0.00%19of 2923of 54Claude Opus 5.518.50 pts
APEX-Agents-AAArtificial Analysis32.01%7of 7no list readGemini 3 5 Flash15.04 pts
τ³-BankingArtificial Analysis21.44%28of 30no list readQwen3.8 Max29.90 pts
ITBench-AAArtificial Analysis30.33%18of 18no list readGPT-5.6 Sol25.88 pts
IFBenchArtificial Analysis77.14%2of 10no list readGrok 4 20 03095.79 pts
AA-AnalystAgentArtificial Analysis41.25%12of 18no list readGemini 3.7 Flash18.75 pts
MLCR-AAArtificial Analysis15.56%30of 37no list readClaude Sonnet 5.559.44 pts
ARC-AGI-3ARC Prize · Standard0.42%8of 11no list readGPT-6 Astra62.29 pts
Vibe Code Bench 1–100Vals AI6.69%20of 2020of 20Claude Opus 5.523.67 pts
τ²-Bench TelecomArtificial Analysis95.61%2of 10no list readJt 35b Flash3.51 pts
Harvey LAB-AAArtificial Analysis58.85%20of 20no list readMuse Spark 1.3 Max36.63 pts
EnterpriseOps-Gym-AAArtificial Analysis42.17%14of 22no list readClaude Fable 58.95 pts
AutomationBench-AAArtificial Analysis35.43%36of 40no list readGemini 4 Argon42.08 pts
SciCodeArtificial Analysis58.68%11of 40no list readClaude Opus 5.58.22 pts
GDP.pdfArtificial Analysis17.80%26of 40no list readGPT-6.1 Sol13.20 pts
CritPtArtificial Analysis17.71%28of 42no list readGPT-5.6 Sol14.58 pts
AA-OmniscienceArtificial Analysis31.8810of 42no list readClaude Opus 5.514.54
AA-LCR v1.1Artificial Analysis82.00%19of 42no list readKimi K36.67 pts
Terminal-Bench 2.1Vals AI70.79%24of 3527of 75Claude Opus 5.516.85 pts
ProofBench v1.1Vals AI26.00%29of 3434of 46Claude Sonnet 5.574.00 pts
LegalBenchVals AI87.40%4of 274of 149Claude Fable 51.16 pts
Harvey's Legal Agent BenchmarkVals AI0.00%37of 3757of 73Muse Spark 1.225.42 pts
MysteryMechanismVals AI20.72%18of 2118of 21GPT-6 Astra32.43 pts
Vals Multimodal IndexVals AI56.07%15of 1720of 33Claude Fable 518.08 pts
BioMysteryBenchVals AI60.00%19of 1920of 21Claude Sonnet 5.521.11 pts
Code MigrationVals AI17.31%34of 3747of 71Claude Sonnet 5.552.52 pts
EMBVals AI52.62%34of 3747of 69Claude Fable 5.124.05 pts
Finance Agent (v2)Vals AI42.98%36of 3752of 73Gemini 4 Argon22.42 pts
IOIVals AI51.83%21of 3122of 38Gemini 4 Argon48.17 pts
Legal Research BenchVals AI20.67%36of 3749of 72Muse Spark 1.3 Max34.62 pts
MedCodeVals AI59.06%2of 352of 104Claude Opus 54.51 pts
MedScribeVals AI76.11%34of 3674of 106Claude Opus 5.515.32 pts
Public Benefits Bench v1.1Vals AI53.79%29of 2938of 45Claude Opus 523.14 pts
SAGEVals AI48.68%14of 3226of 90Claude Opus 4.77.42 pts
Tax Agent BenchVals AI37.84%36of 3749of 64Claude Fable 5.139.80 pts
GDPval-AA v2.1Artificial Analysis775.9841of 42no list readClaude Opus 5.51,070.19
AA-Briefcase v1.1Artificial Analysis453.3541of 41no list readClaude Opus 5.51,368.5
Humanity's Last ExamArtificial Analysis47.03%16of 42no list readClaude Opus 5.514.32 pts

Nearby on Vals Index

The two models above and below on the Vals Index (Vals AI), on every column where this model has a published number.

ModelVals Index%Artificial Analysis Intelligence IndexindexTerminal-Bench 4.0%SWE-bench%LiveCodeBench%GPQA Diamond%MMLU Pro%MMMU Pro%Vibe Code Bench v1.1%ProgramBench%
Grok 4.5xAI44.72%3910.61%86.60%87.35%92.93%89.22%61.79%69.00%0.00%
MiniMax M3MiniMax36.53%292.02%75.00%82.15%92.68%84.22%81.16%47.57%–
Gemini 3.1 ProGoogle33.44%304.04%78.80%88.49%95.45%90.99%88.21%32.03%0.00%
InklingThinking Machines28.67%251.01%77.60%85.52%87.12%86.30%78.50%19.21%0.00%
Bold marks a figure better than this model’s on the same column, whichever model it belongs to. Ranks cover the models shown from the stated source.

Cost and speed, as published

Additional measurements from Artificial Analysis’ own row. Read 30 September 2026.

As publishedCost per taskOutput speed
Gemini 3.1 Pro Preview$0.67113.9 tok/s

Results by benchmark

Independent and vendor results retain their own harnesses and effort settings.

BenchmarkIndependent runAs reported by a publisher
APEX-Agents-AAKnowledge work32.01%Artificial Analysisnot published
τ³-BankingTool use21.44%Artificial Analysisnot published
ITBench-AAIT operations30.33%Artificial Analysisnot published
IFBenchInstruction following77.14%Artificial Analysisnot published
AA-AnalystAgentData analysis41.25%Artificial Analysisnot published
MLCR-AAMedical reasoning15.56%Artificial Analysisnot published
ARC-AGI-3Interactive reasoning0.42%ARC Prize · Standardnot published
Vibe Code Bench 1–100Application development6.69%Vals AInot published
τ²-Bench TelecomTool use95.61%Artificial Analysisnot published
Harvey LAB-AALegal reasoning58.85%Artificial Analysisnot published
EnterpriseOps-Gym-AAEnterprise workflows42.17%Artificial Analysisnot published
AutomationBench-AAWorkflow automation35.43%Artificial Analysisnot published
SciCodeScientific coding58.68%Artificial Analysisnot published
GDP.pdfDocument reasoning17.80%Artificial Analysisnot published
CritPtScientific reasoning17.71%Artificial Analysisnot published
AA-OmniscienceFactual knowledge31.88Artificial Analysisnot published
AA-LCR v1.1Long-context reasoning82.00%Artificial Analysisnot published
Terminal-Bench 4.0Agentic coding4.04%Artificial Analysis
2.52%Vals AI
not published
Terminal-Bench 2.1Agentic coding70.79%Vals AI
73.78%Artificial Analysis
not published
SWE-bench (Verified)Software engineering78.80%Vals AInot published
LiveCodeBenchCode generation88.49%Vals AInot published
GPQA DiamondScience reasoning95.45%Vals AI
94.14%Artificial Analysis
not published
MMLU ProKnowledge90.99%Vals AInot published
MMMU ProMultimodal reasoning88.21%Vals AI
82.43%Artificial Analysis
not published
Vibe Code Bench v1.1Product coding32.03%Vals AInot published
ProgramBenchAgentic coding0.00%Vals AInot published
ProofBench v1.1Mathematics26.00%Vals AInot published
LegalBenchProfessional knowledge87.40%Vals AInot published
Harvey's Legal Agent BenchmarkProfessional knowledge0.00%Vals AInot published
MysteryMechanismScientific reasoning20.72%Vals AInot published
Vals Multimodal IndexComposite56.07%Vals AInot published
Vals IndexComposite33.44%Vals AInot published
BioMysteryBenchAgentic science60.00%Vals AInot published
Code MigrationAgentic coding17.31%Vals AInot published
EMBFinancial modelling52.62%Vals AInot published
Finance Agent (v2)Financial analysis42.98%Vals AInot published
IOICompetitive programming51.83%Vals AInot published
Legal Research BenchProfessional knowledge20.67%Vals AInot published
MedCodeHealthcare administration59.06%Vals AInot published
MedScribeHealthcare administration76.11%Vals AInot published
Public Benefits Bench v1.1Public services53.79%Vals AInot published
SAGEEducation48.68%Vals AInot published
Tax Agent BenchProfessional knowledge37.84%Vals AInot published
GDPval-AA v2.1Knowledge work775.98Artificial Analysisnot published
AA-Briefcase v1.1Knowledge work453.35Artificial Analysisnot published
ChartographyVisual chart recognitionno published result
26.1%Chartography (no tools)Anthropic · 7 October 2026Rank 8 of 10 in this publisher table
Humanity's Last ExamMultidisciplinary reasoning47.03%Artificial Analysisnot published
Artificial Analysis Intelligence IndexComposite30Artificial Analysisnot published
Independent column: the suite named under each figure, one harness per suite. Publisher column: each lab’s own table, reproduced on the benchmark’s page. Publisher ranks compare reported scores in the same table row only; ties share a rank and unreported cells are excluded.

Sources

Read on 17 October 2026.