Atlas

Compare Models

Gemini 3.8 Flash vs GPT-6 Astra

GPT-6 Astra leads by 6.6 capability points (≈95% interval +4.1 to +9.0): a clear lead over Gemini 3.8 Flash · P(higher) ≈ 100%

Wins 253 of 343 shared benchmarks (68 lost, 22 tied)

Capability vs. relative cost
Gemini 3.8 FlashGPT-6 AstraClosed weightsOpen weights

Each line runs through one model’s priced reasoning settings, cheapest to most expensive; grey points are other models’ priced settings in the same range. Results whose reasoning effort the source didn’t report are left out wherever a model has named effort levels. Capability and relative cost are the published estimates (relative cost is cost per task as a multiple of GPT-5 (high)); hover a setting, or use the arrow keys, to set it beside the other model’s same setting.

Gemini 3.8 FlashvsGPT-6 Astracapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins68 – 253 · 343 shared, 22 tied

Comparing matching setups only (127 benchmarks excluded) · suites counted once
← Gemini 3.8 FlashGPT-6 Astra →

Gemini 3.8 FlashvsGPT-6 Astrawins by categoryData Nov 15, 2026Data: Vals AI, Artificial Analysis, benchmarks.bio and 80 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-6 Astra minus Gemini 3.8 Flash.

Clipping Bench v1.0Games5969514+95.9
Short Form Bench v1.0Games19402803+85.5
Concept Synth INDUCTION Challenge100 Round 1General QA9.094.0+85.0
Concept Synth INDUCTION Challenge64 Round 1General QA14.198.4+84.4
BrokenArXiv 06/2026Math20.899.1+78.2
Olam Social Poker EloGames14791550+77.2
FrontierMath Tier 4 (v2) - PrivateMath22.097.6+75.6
BrokenArXiv 05/2026Math19.091.0+72.0
Time Horizon Index: KSPAgents18.890.5+71.7
AA-PersonalityBench Transparency indexChat & Writing23.894.3+71.2
ARC-AGI-3 (Semi-Private)General QA35.099.9+64.9
BrokenArXiv 04/2026Math33.697.5+63.9
ViBench 2.0Code0.060.3+60.3
Terminal-Bench Science 0.1 (Vals)Science8.665.7+57.1
AA-PersonalityBench Slop indexChat & Writing71.416.2−55.8
RSI Arena Stage 1 human ratingAgents9811095+54.8
AA-PersonalityBench Informativeness indexChat & Writing81.029.6−51.9
ProofBench v1.1Math48.099.0+51.0
Terminal-Bench-Science 0.1Science10.059.5+49.5
BARC-Bench (all-safe rate)Safety23.672.7+49.1
Furniture Assembly BenchmarkMultimodal31.780.0+48.3
FrontierSWE V2 Worst@5Code9.555.1+45.6
IOI (Vals v2)Code56.9100.0+43.1
Code2Video-AA EloCode8991427+42.8
Code Migration v1.1Code38.680.6+41.9
FrontierSWE V2 Best@5Code31.673.0+41.4
AA-PersonalityBench Conviction indexChat & Writing84.844.8−40.4
Harvey's Legal Agent Benchmark - Emerging Companies Venture CapitalLegal40.00.0−40.0
MASKSafety58.698.5+39.9
Terminal-Bench 4.0Code19.759.6+39.9
Hedge-BenchFinance & Accounting4.744.2+39.5
TaxCalcBench TY25Finance & Accounting6.044.0+38.0
LLM Creative Story WritingChat & Writing0.13.4+37.3
HPCT Refusal (CAIS)Safety13.050.0+37.0
Mystery Game PuzzlesGames47.084.0+37.0
ArXivMath 05/2026Math58.395.0+36.7
AA-PersonalityBench Imagination indexChat & Writing41.077.2+36.5
ALE-BenchCode21823164+34.8
Terminal-Bench 4.0 (Artificial Analysis)Code19.754.0+34.3
EnigmaEvalGeneral QA37.269.7+32.5
Code MigrationCode36.567.7+31.2
CrossSearch (pass^3)Search29.060.0+31.0
Machiavelli Deception (CAIS)Safety86.576.3+30.9
RareDiseaseBench V1 (35 evaluations)Science2.933.3+30.5
GameBench 2Games64.094.3+30.4
ChartographyMultimodal40.971.0+30.1
ScBench V1 Clean RerunsScience34.063.9+29.9
HoneyBench v0.2 (hack rate)Safety80.550.8+29.7
Vals Science IndexIndexes35.865.1+29.3
AA-3D v1.0 EloCode9741168+28.2
ArXivMath 06/2026Math67.494.4+27.1
ArXivMath 04/2026Math64.290.8+26.7
Humanity's Last Exam DiamondGeneral QA34.360.6+26.3
Artificial Analysis Intelligence Index v5.0.1Indexes30.856.9+26.0
FrontierMath Tiers 1-3 (v2) - PrivateMath68.493.7+25.3
GeneBench-Pro Public Release SubsetScience11.036.0+25.0
Artificial Analysis Intelligence Index v5.0Indexes30.555.3+24.7
Decision-Bench v1.1Professional Work26.951.5+24.6
DAYJOB: AccountingFinance & Accounting6.030.0+24.0
HiL-Bench (Human-in-Loop Benchmark)Agents41.565.3+23.8
MMMU-Pro Extended (Mercor held-out)Multimodal36.460.0+23.6
LMArena Agent Arena Net Improvement (Steering Burden)Agents2.314.5+23.1
MetagenomicsBench V1Science26.749.7+23.0
MetagenomicsBench V0Science26.749.7+23.0
DeepsecBench-AA (Artificial Analysis)Cybersecurity14.536.9+22.4
Creative Writing v3 EloChat & Writing17442177+21.9
LMArena Agent Arena Net ImprovementAgents3.810.4+21.8
MentalHealthBenchSafety35.557.3+21.8
RedlineBenchLegal41.362.8+21.5
FrontierCode 2.0 (Artificial Analysis)Code18.740.1+21.4
AA-PersonalityBench Expressiveness indexChat & Writing79.158.1−21.2
RiemannBenchMath51.272.0+20.8
PLE (Physical-Engineering's Last Exam)Professional Work27.648.3+20.7
ORCA-bench Hard RCA (Fireworks 324-task subset)Agents29.550.2+20.7
GBENCH One-shot Coding GScoreGames58.679.2+20.6
sudo L7 (October 2026 rescoring)Code10.631.1+20.5
VIALSMultimodal29.449.9+20.5
TextQuests Game ProgressGames54.174.2+20.1
Harvey's Legal Agent Benchmark - Banking FinanceLegal20.00.0−20.0
Harvey's Legal Agent Benchmark - Capital MarketsLegal20.00.0−20.0
Harvey's Legal Agent Benchmark - Corporate GovernanceLegal20.00.0−20.0
Harvey's Legal Agent Benchmark - Corporate M&ALegal0.020.0+20.0
Harvey's Legal Agent Benchmark - Energy Natural ResourcesLegal20.00.0−20.0
Harvey's Legal Agent Benchmark - Healthcare Life SciencesLegal20.00.0−20.0
Harvey's Legal Agent Benchmark - Structured Finance SecuritizationLegal20.00.0−20.0
HardGeoBench Avg DistanceMultimodal8431478−19.6
SheetBench-50Finance & Accounting45.664.7+19.1
LiveBench 2.0Agents67.386.1+18.8
DAYJOB: FinanceFinance & Accounting2.821.5+18.7
CritPt v2 (agentic, avg@5)Science65.183.7+18.6
MaxxBenchCode53.471.9+18.5
NoCoT-Bench NCRI 15.2General QA126159+18.3
Hermes IndexIndexes38.156.3+18.1
SciUniverse Level 1Science14.632.5+17.8
TxBench-Antibody-Development V1 (60 evaluations)Science15.032.8+17.8
Physical-Engineering's Last Exam (hidden test)Professional Work28.245.8+17.6
FrontierSWE V2 (Mercor rerun, mean@3)Code12.829.9+17.1
BioMysteryBench (Vals Terminus 2)Science62.279.3+17.0
WellnessBenchSafety47.064.0+17.0
AA-Briefcase EloAgents12021494+17.0
MysteryMechanismScience36.553.2+16.7
Web Clone BenchCode1.017.5+16.5
TissueMicroscopyBench V1 (49 evaluations)Science3.419.7+16.3
TextQuests Average Cumulative HarmSafety24.429.2−16.3
Agent Red Teaming k=100 (CAIS)Safety29.813.5+16.3
TxBench Oligo Discovery V1 Clean RerunsScience33.149.2+16.1
Chess Puzzles (task version 2.0)Games58.074.0+16.0
EnterpriseBench: CoreCraftProfessional Work58.542.6−15.9
Design Creativity BenchCode60.665.5+15.5
AA-PersonalityBench Warmth indexChat & Writing60.075.3+15.4
VariantBench V1 Clean RerunsScience31.446.6+15.3
HardGeoBench Country AccuracyMultimodal67.552.5−15.0
MBCT (SecureBio)Science48.963.9+15.0
Snorkel Agentic Coding 2.0Code32.647.6+15.0
SpatialViz-BenchMultimodal78.193.1+15.0
Remote Labor Index (RLI)Professional Work5.820.8+15.0
MKA (Artificial Analysis)Professional Work35.350.2+15.0
IntPhys 2Multimodal62.577.4+14.9
HieroglyphBenchMultimodal54.640.0−14.6
Finance Agent v2 - ComparablesFinance & Accounting52.037.8−14.2
HardGeoBench <=100miMultimodal60.046.3−13.7
ChessBench EloGames15271755+13.6
CritPt (avg@5)Science18.331.7+13.4
LMArena Agent Arena Praise vs ComplaintAgents4.931.7+13.4
BabaIsBench Stage 1: The LakeGames86.7100.0+13.3
Medical Long Context Reasoning (MLCR-AA)Health & Medicine21.735.0+13.3
RT-SAFE Collisions per EpisodeSafety27.722.4+13.3
Document ArenaProfessional Work14821467−13.1
LLM Debate BenchmarkChat & Writing15141606+13.0
Finance Agent v2 - Market AnalysisFinance & Accounting78.565.5−13.0
VCT v2 (SecureBio)Science45.258.0+12.8
BioSecBench-Function V1 Clean RerunsScience36.048.6+12.7
APEX-SWECode36.448.6+12.2
SWE Atlas v1.1 - OverallCode57.369.4+12.1
Humanity's Sixth SenseMultimodal41.653.6+12.0
Arena Alignment IndexSafety75.887.8+12.0
Finance Agent v2 - Earnings AnalysisFinance & Accounting81.469.5−11.9
Artificial Analysis Intelligence Index v4.3.2Indexes40.952.7+11.7
AutomationBench 1.0.6 (Zapier)Agents29.741.4+11.7
AutoResearchExam (hidden-test AUARC)Agents0.50.6+11.7
Artificial Analysis Intelligence Index v4.3Indexes41.252.8+11.6
VCT (SecureBio)Science46.958.5+11.6
Hermes BenchAgents62.273.7+11.5
LMCAGeneral QA52.964.4+11.5
SAGEMath35.146.4+11.3
Conceptual Reasoning IndexIndexes65.773.9+11.3
VCT-Refusal (CAIS)Safety6.417.6+11.2
Visa Agentic Commerce BenchmarkAgents53.364.4+11.1
BrowseComp (Mercor 130-question subset)Search83.194.2+11.1
Blueprint-Bench 2Multimodal0.40.5+11.1
Finance Agent v2 - Financial ModelingFinance & Accounting32.621.6−11.1
FrontierCode 1.1 Extended - ScoreCode53.564.5+11.1
Shroombench (mushroom identification, top-1)Multimodal65.054.0−11.0
CWE-Bench-AA (Artificial Analysis)Cybersecurity52.563.3+10.8
DAYJOB: HealthcareHealth & Medicine0.811.6+10.8
SpatialBench-Long V1 Clean RerunsScience28.639.3+10.7
HealthBench ProfessionalHealth & Medicine59.069.7+10.7
SM BenchSafety80.469.9−10.5
micro1 PharmaBenchHealth & Medicine5.616.1+10.5
Artificial Analysis Progress Index v1Indexes320350+10.3
HardGeoBench <=10miMultimodal46.336.3−10.0
HealthBench (length-adjusted, Artificial Analysis)Health & Medicine48.458.2+9.8
MA-BenchProfessional Work43.653.4+9.8
DAXBench v2Code87.296.8+9.6
Finance Agent v2 - All PassFinance & Accounting49.740.3−9.4
Ivo-micro1 ContractBenchLegal55.045.6−9.4
GDP.pdf (Artificial Analysis, All-pass)Professional Work22.832.2+9.4
ComplexConstraintsChat & Writing48.457.7+9.3
PCT Political Inconsistency (CAIS)Safety46.337.4+8.9
Vibe Code Bench 1–100Code18.827.6+8.9
Epoch Capabilities Index (ECI)Indexes159169+8.5
WeirdML (v2)Science84.893.3+8.5
CyberGym-E2E-AA (Artificial Analysis)Cybersecurity8.40.0−8.4
Artificial Analysis Cyber Index v1Indexes25.133.4+8.3
Quant-Arena (iterative total score)Finance & Accounting43.051.3+8.3
SpatialBench V1 Clean RerunsScience59.467.5+8.1
LMArena Agent Arena Steering BurdenAgents-1.314.8+8.0
OncologyBench Strict (25 tasks, October 9 rubric)Health & Medicine13.321.3+8.0
Toloka Arena (v3)Agents59.067.0+8.0
Bio Capabilities Index (SecureBio)Indexes153164+7.9
Build-A-Chip-BenchAgents0.88.7+7.9
SUPERChemScience72.680.3+7.7
AutomationBench-AA 1.0.6Agents60.968.5+7.6
MissionReportProfessional Work54.361.8+7.5
Humanity's Last ExamGeneral QA46.253.6+7.4
ScBench-Long V1 Clean RerunsScience19.927.3+7.4
MacroscopeBenchCode70.778.0+7.3
NoCoT-Bench NCKI (kspine v3)General QA121131+7.3
SWE Atlas - Codebase QnACode40.647.8+7.2
AA-Omniscience IndexIndexes29.643.7+7.1
NeuroscienceBench V1 (91 evaluations)Science15.822.7+7.0
EpiBench V1 Clean RerunsScience21.428.3+6.9
Bedside BenchHealth & Medicine86.593.4+6.9
BuseyBench SVGMultimodal6.67.3+6.9
Finance Agent v2 - Disclosure AnalysisFinance & Accounting68.561.7−6.8
SpeciEval (Refusal-Aware Retries, Historical)Safety89.996.5+6.7
SpeciEval (Pooled English Runs)Safety89.996.5+6.7
ATLAS-Finance v1.0Finance & Accounting4.711.3+6.7
BioSecBench-Refusal V1 Clean RerunsSafety47.440.8−6.6
HPCT (SecureBio)Science62.469.0+6.6
Finance Agent v2 - AdjustmentsFinance & Accounting54.848.3−6.5
LMArena Agent Arena Confirmed SuccessAgents1.314.0+6.4
FACTS Benchmark SuiteGeneral QA69.475.7+6.3
SnorkelFinance 2.0Finance & Accounting5.411.6+6.2
APEX-AccountingFinance & Accounting51.757.9+6.2
CharXiv Extended (Mercor held-out)Multimodal78.985.1+6.2
Snorkel Digital Work IndexIndexes16.923.0+6.1
Snorkel SWE-bench CLICode6.312.4+6.1
BioSecBench-RespiratoryVirus-Surveillance V1 (110 evaluations)Science7.313.3+6.1
ATLAS-Finance v1.1Finance & Accounting3.79.7+6.0
Eval Connections One-Shot with TrapsGames88.094.0+6.0
Longform Creative WritingChat & Writing76.882.8+6.0
SimpleQA Verified (Epoch anti-abstention)General QA69.775.6+5.9
Finance Agent v2 - PrecedentsFinance & Accounting36.130.3−5.9
ARC-AGI-2General QA89.295.0+5.8
TxBench Antibody Discovery V1 Clean RerunsScience43.348.9+5.6
SnorkelRevOpsProfessional Work5.410.9+5.5
WCB (SecureBio)Science57.162.6+5.5
SciCode (main problems)Science14.219.6+5.4
Finance Agent v2.1Finance & Accounting65.860.4−5.3
LAB-Bench ProtocolQA (SecureBio)Science84.089.3+5.3
HLAB-Vals-1.1Legal12.17.1−5.0
AA-AnalystAgent (pass^5)Professional Work46.351.3+5.0
HANDBOOK.md Agents v1.1Professional Work26.531.4+4.9
Vals Finance IndexIndexes60.555.6−4.8
Westworld Finance Diligence BenchFinance & Accounting49.754.5+4.8
TxBench Preclinical Pharmacology V1 Clean RerunsScience54.759.3+4.7
GDP.xlsxProfessional Work18.322.9+4.6
ProgramBenchCode1.05.5+4.5
BioSecBench-Surveillance V1 Clean RerunsScience44.148.5+4.4
SnorkelLegalLegal25.320.9−4.4
BioSecBench-Surveillance V0 (102 tasks)Science44.148.5+4.4
Epiq LAER CorporateBenchProfessional Work77.682.0+4.4
Vals Index v2Indexes62.366.6+4.4
Immigration BenchLegal63.467.7+4.3
GDPval-AA v2.1Professional Work14421538+4.2
CVDP (517-problem set)Code50.754.9+4.2
ChessBench AccuracyGames81.185.0+4.2
Last Translation Benchmark (LTBv1-eval, blind)Chat & Writing39.844.0+4.2
LMArena Text Arena EloChat & Writing14971474−4.1
AA-LCR v1.1Search84.080.0−4.0
BioTIER RefusalSafety83.979.9−4.0
CadQueryEval (Corrected Geometry)Code96.0100.0+4.0
DAYJOB: TaxFinance & Accounting0.84.8+4.0
ITBench-AAAgents52.548.6−4.0
AA-PersonalityBench Organization indexChat & Writing73.369.5−3.8
GDPval Extended (Mercor held-out rubric score)Professional Work60.464.1+3.7
Roboflow Vision Evals (October 2026, five tasks)Multimodal79.883.4+3.6
SWE-TogetherCode64.260.6−3.6
DistressBenchSafety70.774.3+3.6
ACCoRDGeneral QA78.281.7+3.5
Tax Agent BenchFinance & Accounting66.863.3−3.4
MedScribeHealth & Medicine84.587.9+3.4
ChessBenchGames73.370.0−3.3
AIME 2025Math96.7100.0+3.3
LAB-Bench LitQA2 (SecureBio)Science51.554.8+3.3
GDPval-AA v2Professional Work14641530+3.2
GDPval (Mercor 213-task rubric score)Professional Work83.586.7+3.2
Snorkel WorkplaceAgentsProfessional Work11.214.4+3.2
SpatialBench V0 VerifiedScience67.070.1+3.2
LMArena Agent Arena Bash RecoveryAgents-0.75.5+3.1
Gray Swan IPI Attack Success (k=15)Safety5.58.5−3.0
FrontierMath ErdősMath0.02.9+2.9
Ramp Accounting BenchFinance & Accounting45.748.7+2.9
SnorkelManufacturingProfessional Work7.110.0+2.9
LMArena Agent Arena SteerabilityAgents4.3-1.4−2.8
LLM Chess EloGames15461614+2.8
Harvey HLAB — Criteria Pass RateLegal90.287.5−2.7
ERQAMultimodal79.682.3+2.7
𝜏³-BankingAgents45.843.1−2.7
CyberBench v1.1 OverallCybersecurity43.841.1−2.7
Dual-Arm Sim (task completion)Agents0.02.5+2.5
OpenDS-WorldSearch36.539.0+2.5
SWE-bench Multimodal (Vals v1, 480 Tasks)Code40.843.3+2.5
Vision Arena OverallMultimodal12931283−2.5
ChessBench CoherenceGames94.697.0+2.4
ForecastBench Preliminary Dataset Brier IndexForecasting60.562.9+2.4
Terminal-Bench 2.1Code87.689.9+2.2
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal11.49.2−2.2
APEX-Agents 1.1Professional Work72.975.1+2.2
SnorkelUnderwrite 2.0Finance & Accounting19.121.2+2.1
MedXpertQA MM (200-question set)Health & Medicine84.786.8+2.1
Finance Agent v2 - General Quantitative AnalysisFinance & Accounting80.978.8−2.1
SWE-Bench Pro V2 — FullCode94.996.9+2.0
DualEntry Accounting AI BenchmarkFinance & Accounting83.281.2−2.0
Professional Reasoning Benchmark - FinanceFinance & Accounting49.547.5−2.0
MMMU ProMultimodal84.586.4+1.9
Creative Writing v3 Rubric ScoreChat & Writing15.615.9+1.7
DTBenchGeneral QA95.797.3+1.6
Roboflow Vision Evals (September 2026 protocol)Multimodal85.186.6+1.5
WMDP-Bio (SecureBio)Science89.490.8+1.4
Vals Legal IndexIndexes69.868.4−1.4
Financial Coaching BenchFinance & Accounting66.367.5+1.3
SimpleBenchGeneral QA82.483.6+1.2
OTIS Mock AIME 2024-2025Math98.9100.0+1.1
LMArena Agent Arena Bash ReliabilityAgents9.97.8−1.0
GPQA Diamond (Artificial Analysis avg@5)General QA95.396.3+1.0
Harvey LAB: In-House Contracting (held-out)Legal6.35.3−1.0
MMLU ProGeneral QA90.791.5+0.8
CharXiv (descriptive and reasoning combined)Multimodal95.496.1+0.7
Extended NYT Connections 940 (Quadratic v1)Games97.498.1+0.7
BioSimBench 1.0Science15.315.9+0.6
ScarfBench Smoke ScoreCode66.065.4−0.6
DeepSWE v1.1Code73.873.2−0.6
Professional Reasoning Benchmark - LegalLegal48.948.4−0.5
Prophet Arena Arena Score (Fixed Context)Forecasting85.384.7−0.5
APEX-1 (Mercor, mean score)Professional Work68.869.3+0.5
MindCube TinyMultimodal92.292.7+0.5
Excel Modeling BenchmarkFinance & Accounting72.271.7−0.5
Legal Research BenchLegal38.939.4+0.5
SEA-HELMGeneral QA77.376.9−0.4
LVBenchMultimodal87.187.5+0.4
Qodo AI Code Review Benchmark (125 PRs)Code47.046.6−0.4
Finance Agent v2 - General Qualitative AnalysisFinance & Accounting68.168.4+0.4
MedCodeHealth & Medicine48.148.5+0.4
LMArena Agent Arena Tool HallucinationAgents0.41.0+0.3
HealthBenchHealth & Medicine57.156.9−0.2
Excel Modeling Benchmark v1.1Finance & Accounting72.972.7−0.2
Prophet Arena Arena Score (Agentic, OpenCode)Forecasting86.086.2+0.2
SciCode 1.0.1 (Artificial Analysis)Code56.656.5−0.1
LegalReasonBenchLegal39.639.5−0.1
ARC-AGI-1 Semi-PrivateGeneral QA98.598.50.0
BabaIsBench Stage 0: The IntroGames100.0100.00.0
CloudOpsBench v1.0Agents42.842.80.0
Harvey's Legal Agent Benchmark - Antitrust CompetitionLegal0.00.00.0
Harvey's Legal Agent Benchmark - Arbitration International Dispute ResolutionLegal0.00.00.0
Harvey's Legal Agent Benchmark - Bankruptcy RestructuringLegal0.00.00.0
Harvey's Legal Agent Benchmark - Data Privacy CybersecurityLegal0.00.00.0
Harvey's Legal Agent Benchmark - Employment LaborLegal0.00.00.0
Harvey's Legal Agent Benchmark - Environmental ESGLegal0.00.00.0
Harvey's Legal Agent Benchmark - Funds Asset ManagementLegal0.00.00.0
Harvey's Legal Agent Benchmark - ImmigrationLegal0.00.00.0
Harvey's Legal Agent Benchmark - InsuranceLegal0.00.00.0
Harvey's Legal Agent Benchmark - Intellectual PropertyLegal0.00.00.0
Harvey's Legal Agent Benchmark - International Trade SanctionsLegal0.00.00.0
Harvey's Legal Agent Benchmark - Litigation Dispute ResolutionLegal0.00.00.0
Harvey's Legal Agent Benchmark - Real EstateLegal0.00.00.0
Harvey's Legal Agent Benchmark - TaxLegal0.00.00.0
Harvey's Legal Agent Benchmark - Trusts Estates Private ClientLegal20.020.00.0
Harvey's Legal Agent Benchmark - White Collar Defense InvestigationsLegal0.00.00.0
LAB-Bench CloningScenarios (SecureBio)Science100.0100.00.0
MedXpertQA MM Extended (Mercor held-out)Health & Medicine60.860.80.0
ObviousBenchGeneral QA100.0100.00.0

Gemini 3.8 FlashvsGPT-6 Astranormalised score marginsData Nov 15, 2026Data: Vals AI, Artificial Analysis, benchmarks.bio and 80 othersVillage LLM Atlasatlas.village.dev

Win counts use each benchmark’s declared score direction. Represented suite slices and the derived Capability score are excluded from headline totals. Missing or incompatible setups are excluded, never counted as losses.
Loading Atlas data…