OverviewCompare models

Compare models

Two releases, one evaluator field and one scale. Missing results stay missing.

Compare one test

Choose a test and one evaluator field.

aaArtificial Analysis% · Higher is better
  1. 1Claude Opus 5.559.60%
  2. 2Gemini 3.8 Flash19.70%

Bars start at zero. 2 reported results for the selected models; ranks apply to this selection only; ties share a rank. Published effort settings may vary. Source ↗

All shared tests

Same evaluator and unit. Percentage differences are percentage points.

Show all 57 shared tests Same evaluator · same unit
Test / evaluatorGemini 3.8 FlashClaude Opus 5.5DifferenceModel 1 − model 2
τ³-Bankingaa% · Higher is better44.95No result—
ITBench-AAaa% · Higher is better52.5438.23+14.31 pp
MLCR-AAaa% · Higher is better21.6766.67-45.00 pp
ARC-AGI-3arc prize · standard% · Higher is better10.37No result—
ARC-AGI-3arc prize · provider adapter% · Higher is better35.00No result—
SWE-Bench Pro V2 Fullscale ai / seal% · Higher is better94.86No result—
SWE-Bench Pro V2 HARDscale ai / seal% · Higher is better58.80No result—
Vals RSI Indexvals% · Higher is better19.9337.31-17.38 pp
Vibe Code Bench 1–100vals% · Higher is better18.7730.36-11.59 pp
CUA-benchvals% · Higher is better4.1714.00-9.83 pp
Harvey LAB-AAaa% · Higher is betterNo result91.20—
AutomationBench-AAaa% · Higher is better59.9369.54-9.61 pp
SciCodeaa% · Higher is better56.6066.90-10.30 pp
GDP.pdfaa% · Higher is better21.0026.20-5.20 pp
CritPtaa% · Higher is better18.2931.71-13.42 pp
AA-Omniscienceaascore · Higher is better29.5546.42-16.87 score
AA-LCR v1.1aa% · Higher is better81.3384.67-3.34 pp
Terminal-Bench 4.0aa% · Higher is better19.7059.60-39.90 pp
Terminal-Bench 4.0vals% · Higher is better19.1965.15-45.96 pp
Terminal-Bench 2.1vals% · Higher is better81.2787.64-6.37 pp
Terminal-Bench 2.1aa% · Higher is better87.64No result—
SWE-bench (Verified)vals% · Higher is better80.00No result—
LiveCodeBenchvals% · Higher is better89.48No result—
GPQA Diamondvals% · Higher is better94.44No result—
GPQA Diamondaa% · Higher is better95.25No result—
MMLU Provals% · Higher is better90.22No result—
MMMU Provals% · Higher is better89.08No result—
MMMU Proaa% · Higher is better85.6187.69-2.08 pp
Vibe Code Bench v1.1vals% · Higher is better78.6590.29-11.64 pp
Terminal-Bench Science 0.1aa% · Higher is better10.0059.05-49.05 pp
Terminal-Bench Science 0.1vals% · Higher is better8.5747.14-38.57 pp
ProgramBenchvals% · Higher is better1.0018.50-17.50 pp
ProofBench v1.1vals% · Higher is better48.00100.00-52.00 pp
LegalBenchvals% · Higher is better86.99No result—
Harvey's Legal Agent Benchmarkvals% · Higher is better10.003.75+6.25 pp
SkillsBenchvals% · Higher is better57.99No result—
MysteryMechanismvals% · Higher is better36.4949.55-13.06 pp
Vals Indexvals% · Higher is better54.8366.97-12.14 pp
BioMysteryBenchvals% · Higher is better62.2279.26-17.04 pp
Code Migrationvals% · Higher is better36.5566.65-30.10 pp
CyberBench v1.1vals% · Higher is better43.7555.36-11.61 pp
EMBvals% · Higher is better72.2075.94-3.74 pp
Finance Agent (v2)vals% · Higher is better61.4458.59+2.85 pp
IOIvals% · Higher is better56.9495.06-38.12 pp
Legal Research Benchvals% · Higher is better38.9450.48-11.54 pp
MedCodevals% · Higher is better48.1349.80-1.67 pp
MedScribevals% · Higher is better84.5091.43-6.93 pp
Public Benefits Bench v1.1vals% · Higher is better65.2970.64-5.35 pp
SAGEvals% · Higher is better35.0645.83-10.77 pp
SRE Benchvals% · Higher is betterNo result33.59—
Tax Agent Benchvals% · Higher is better66.7770.50-3.73 pp
DeepSWE v1.1datacurve% · Higher is better73.83No result—
FrontierCode v1.1aa% · Higher is better36.3652.17-15.81 pp
GDPval-AA v2.1aaElo · Higher is better1,411.951,846.17-434.22 Elo
AA-Briefcase v1.1aaElo · Higher is better1,201.591,821.85-620.26 Elo
Humanity's Last Examaa% · Higher is better47.8261.35-13.53 pp
Artificial Analysis Intelligence Indexaaindex · Higher is better4158-17.00 index