benchgap
Google · model

Gemini 4 Argon benchmark scores

As of 2026-11-21, Gemini 4 Argon (Google) has measured scores on 55 benchmarks and estimated scores on 107 more.

BenchmarkScoreSource
AA-GPQA Diamond96.6%estimated ± 1.2 pp, low confidence
AA-HLE57.1%measured
AA-SciCode61.8%measured
AIME2698.0%estimated ± 1.4 pp, medium confidence
Artificial Analysis Intelligence Index50.4%measured
CritPt27.1%measured
GDPval-AA56.2%measured
GPQA Diamond96.4%estimated ± 1.4 pp, high confidence
HLE67.5%estimated ± 2.2 pp, high confidence
AA-LCR79.7%measured
LiveCodeBench v6100.0%estimated ± 4.9 pp, medium confidence
AA-Omniscience Accuracy48.3%measured
SciCode61.3%estimated ± 3.1 pp, medium confidence
Terminal-Bench 2.188.7%estimated ± 5.6 pp, low confidence
BrowseComp91.7%estimated ± 1.9 pp, high confidence
HLE w/ tools61.0%estimated ± 5.3 pp, medium confidence
LongBench v265.4%estimated ± 5.5 pp, medium confidence
AA Coding Index82.3%estimated ± 1.2 pp, medium confidence
AA-MMMU-Pro86.4%estimated ± 3.0 pp, medium confidence
APEX-Agents50.1%estimated ± 7.4 pp, low confidence
APEX-Agents-AA38.9%estimated ± 2.9 pp, low confidence
BioMysteryBench (human-difficult)49.3%estimated ± 6.8 pp, low confidence
DeepSearchQA89.6%estimated ± 9.6 pp, medium confidence
SWE-bench Verified90.2%estimated ± 4.5 pp, medium confidence
AutomationBench51.3%measured
BFCL v477.1%estimated ± 2.7 pp, medium confidence
CyberGym87.4%estimated ± 4.6 pp, medium confidence
JobBench65.7%estimated ± 4.9 pp, low confidence
SWE-bench Pro60.2%estimated ± 2.9 pp, low confidence
WideResearch84.5%estimated ± 7.8 pp, low confidence
AA Agentic Index52.4%estimated ± 3.1 pp, high confidence
MMLU-Pro85.5%estimated ± 5.0 pp, medium confidence
FailureSensorIQ68.7%estimated ± 1.7 pp, medium confidence
GPQA95.5%estimated ± 0.9 pp, high confidence
MMLU90.5%estimated ± 0.9 pp, medium confidence
Vals GPQA Diamond94.1%estimated ± 1.4 pp, medium confidence
Vals LiveCodeBench90.8%estimated ± 3.0 pp, medium confidence
Vals MMLU-Pro92.8%estimated ± 1.4 pp, low confidence
MMMLU92.7%estimated ± 2.6 pp, low confidence
Chess Puzzles59.0%measured
AA MMLU-Pro88.4%estimated ± 0.8 pp, high confidence
Gert Labs74.2%estimated ± 8.5 pp, low confidence
AA-AnalystAgent48.8%measured
AA Harvey LAB v1.093.3%estimated ± 1.3 pp, medium confidence
AA Tau3 Banking50.0%estimated ± 3.3 pp, low confidence
AA Terminal-Bench 2.180.4%estimated ± 2.5 pp, medium confidence
ApprenticeBench44.9%estimated ± 12.5 pp, low confidence
ARC-AGI-198.5%measured
ARC-AGI-291.3%measured
BabaIsBench100.0%estimated ± 1.2 pp, high confidence
CursorBench 3.173.1%estimated ± 3.4 pp, low confidence
CursorBench 3.273.9%estimated ± 2.0 pp, medium confidence
EBR-bench64.8%measured
Terminal-Bench 3.043.6%estimated ± 3.2 pp, medium confidence
FrontierCode 1.1 Main53.8%estimated ± 2.8 pp, medium confidence
FrontierMath v2 (Tier 4)78.0%measured
FrontierMath v2 (Tiers 1-3)86.0%measured
FrontierSWE v255.0%measured
Furniture Assembly72.5%measured
MaxxBench72.1%estimated ± 4.3 pp, medium confidence
MazeBench14.0%measured
MirrorCode58.9%measured
Mystery Game Puzzles64.0%measured
NYT Connections98.6%measured
OSWorld-Verified85.2%estimated ± 8.4 pp, low confidence
ParseBench-177.5%estimated ± 3.7 pp, high confidence
Terminal-Bench 4.155.5%measured
Terminal-Bench Hard67.9%estimated ± 2.8 pp, low confidence
Vals SWE-bench100.0%estimated ± 3.7 pp, medium confidence
Terminal-Bench 2.1 (Vals AI)57.7%measured
VulcanBench v387.6%estimated ± 3.9 pp, medium confidence
CWE-Bench-AA63.3%estimated ± 3.8 pp, high confidence
AA Cyber Index46.3%estimated ± 9.4 pp, medium confidence
DeepsecBench-AA45.4%estimated ± 7.5 pp, low confidence
Harvey LAB-AA v1.17.4%estimated ± 4.0 pp, low confidence
HealthBench Professional (length-adjusted, Artificial Analysis)61.2%estimated ± 4.3 pp, medium confidence
HealthBench Professional (raw, Artificial Analysis)68.9%estimated ± 0.8 pp, medium confidence
AA ITBench47.7%estimated ± 5.1 pp, low confidence
Bug Hunt Bench40.7%measured
Chartography (no tools)71.6%measured
Chartography (tools)91.2%estimated ± 2.7 pp, medium confidence
CursorBench 4.052.9%measured
CWE-bench v168.0%measured
DeepSWE77.9%measured
DRACO87.9%estimated ± 0.7 pp, low confidence
FrontierCode 2.027.6%measured
FrontierPhysics8.3%measured
HLE w/o tools56.2%estimated ± 4.9 pp, high confidence
OSWorld 2.069.2%measured
OSWorld 2.178.8%estimated ± 5.3 pp, medium confidence
OSWorld 2.1 Offline74.7%measured
PostTrainBench v1.145.3%measured
ProgramBench90.9%estimated ± 4.8 pp, medium confidence
SpireClimb30.0%estimated ± 5.4 pp, low confidence
StealthMark67.4%measured
SWE-bench Multimodal74.1%estimated ± 8.3 pp, low confidence
Terminal-Bench 4.057.4%measured
Terminal-Bench-Science 0.151.3%estimated ± 10.7 pp, low confidence
Toolathlon-Verified77.8%estimated ± 0.8 pp, medium confidence
VulcanBench Frontier v492.0%estimated ± 1.8 pp, high confidence
AA AutomationBench77.5%measured
GDP.pdf27.6%estimated ± 3.1 pp, high confidence
AA Terminal-Bench 4.055.1%estimated ± 6.2 pp, medium confidence
ARC-AGI-346.3%measured
ARC-AGI-3 Provider Adapter97.3%measured
cursorBench4156.5%measured
CWE-bench65.0%measured
MKA51.3%measured
SkillsBench official leaderboard (BenchFlow)58.7%estimated ± 6.1 pp, low confidence
BinaryAudit35.8%estimated ± 12.1 pp, low confidence
OTelBench22.5%estimated ± 6.6 pp, low confidence
Vibe Code Bench91.9%measured
FACTS Multimodal54.2%measured
Hermes Bench74.9%estimated ± 1.5 pp, medium confidence
Hermes Index58.9%estimated ± 3.1 pp, high confidence
ParseBench-285.3%measured
CharXiv w/o tools87.6%estimated ± 2.5 pp, low confidence
ExploitBench99.8%estimated ± 1.9 pp, medium confidence
ExploitGym39.5%measured
HMMT Feb 2026100.0%estimated ± 3.1 pp, medium confidence
HMMT Nov 202596.0%estimated ± 1.5 pp, low confidence
MCP Atlas83.5%estimated ± 10.4 pp, low confidence
MMAnswerBench95.7%estimated ± 2.4 pp, low confidence
MMLU-Redux97.0%estimated ± 0.8 pp, low confidence
MMMU-Pro87.7%estimated ± 5.3 pp, low confidence
NL2Repo64.2%estimated ± 6.0 pp, low confidence
PostTrainBench v144.2%estimated ± 1.3 pp, low confidence
SuperGPQA96.3%estimated ± 9.3 pp, low confidence
Terminal-Bench 2.064.6%estimated ± 5.8 pp, medium confidence
Toolathlon55.9%estimated ± 6.3 pp, low confidence
HealthBench Hard33.6%estimated ± 8.6 pp, low confidence
React Native Evals86.5%estimated ± 3.0 pp, low confidence
ResearchClawBench21.9%estimated ± 2.2 pp, low confidence
SWE-Rebench62.3%estimated ± 6.5 pp, low confidence
PostTrainBench40.4%measured
DAYJOB15.7%measured
DAYJOB Healthcare11.6%measured
Finance Agent v265.4%measured
BioMysteryBench (human-solvable)89.5%estimated ± 0.6 pp, medium confidence
FrontierCode 1.1 Extended62.0%estimated ± 3.0 pp, medium confidence
HealthBench (raw)63.0%estimated ± 6.0 pp, medium confidence
HealthBench (length-adjusted)58.4%estimated ± 0.9 pp, medium confidence
HealthBench Professional64.8%estimated ± 2.1 pp, medium confidence
HealthBench Professional (raw)66.6%estimated ± 2.3 pp, medium confidence
HLE-Verified54.8%estimated ± 1.2 pp, low confidence
LABBench288.8%measured
ScarfBench75.2%measured
BlenderWorld (text-only)84.2%estimated ± 2.8 pp, medium confidence
Cookie-Bench90.2%estimated ± 1.3 pp, medium confidence
STEMBench2.4%measured
Agents' Last Exam39.5%measured
OpenHarmony Bench v1.058.9%estimated ± 2.4 pp, medium confidence
OpenHarmony Bench v1.263.8%estimated ± 11.0 pp, low confidence
SEC-Bench Pro83.4%measured
MRCRv285.7%estimated ± 11.3 pp, low confidence
LVBench91.7%measured
MMLongBench-Doc61.4%estimated ± 2.6 pp, low confidence
SWE-Marathon56.6%estimated ± 2.1 pp, medium confidence
BioBasicsBench68.6%estimated ± 2.8 pp, low confidence
LifeVal79.4%estimated ± 2.3 pp, medium confidence
ExploitBench Internal Port6.7%estimated ± 1.4 pp, medium confidence
SWE-Atlas Codebase QnA62.1%estimated ± 4.8 pp, medium confidence