Atlas

Compare Models

GPT-5.5 vs GPT-6 Astra

GPT-6 Astra leads by 7.2 capability points (≈95% interval +4.9 to +9.6): a clear lead over GPT-5.5 · P(higher) ≈ 100%

Wins 209 of 264 shared benchmarks (28 lost, 27 tied)

Capability vs. relative cost
GPT-5.5GPT-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.

GPT-5.5vsGPT-6 Astracapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins28 – 209 · 264 shared, 27 tied

Comparing matching setups only (14 benchmarks excluded) · suites counted once
← GPT-5.5GPT-6 Astra →

GPT-5.5vsGPT-6 Astrawins by categoryData Nov 15, 2026Data: Vals AI, benchmarks.bio, Artificial Analysis and 67 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 GPT-5.5.

Time Horizon Index: KSPAgents8.390.5+82.2
ActiveVisionMultimodal10.676.5+65.9
Clocktower BenchGames14861892+58.2
ARC-AGI-3 (Semi-Private)General QA0.454.8+54.4
SRE Bench — Fully SolvedCode3.856.9+53.1
ExploitBench - release-reported exploit progressCybersecurity47.9100.0+52.1
VoxelBench Text RatingGames19252625+49.9
TasteVal (compute multiplier)Science0.51.6+49.7
Terminal-Bench 4.0 (Artificial Analysis)Code14.659.6+44.9
BrokenArXiv 05/2026Math50.091.0+41.0
ChartographyMultimodal31.071.0+40.0
GSO Opt@1Code40.279.4+39.2
ALE-BenchCode21083164+37.4
MirrorCode (ML, +Private, 2L)Code10.046.7+36.7
ChessBench — Historical Output RepairGames29.366.0+36.7
Humanity's Last Exam DiamondGeneral QA27.963.8+35.9
Furniture Assembly BenchmarkMultimodal44.280.0+35.8
Code2Video-AA EloCode10001427+34.6
GameBench 2Games59.994.3+34.4
LMArena Code Arena WebDevCode15121764+33.0
EnigmaEvalGeneral QA37.269.7+32.5
NoCoT-Bench NCRI 15.2General QA100159+32.3
Olam Social Poker EloGames15211550+31.5
Drone-BenchAgents63.995.1+31.2
BrokenArXiv 06/2026Math69.499.1+29.6
ClockBench AccuracyMultimodal46.175.0+28.9
Agent Red Teaming k=100 (CAIS)Safety41.513.5+28.0
GeneBench-Pro Public Release SubsetScience8.036.0+28.0
Mystery Game PuzzlesGames56.084.0+28.0
TextQuests Average Cumulative HarmSafety21.429.2−26.5
ZeroBench Main Mean pass@1 (five samples)Multimodal17.243.2+26.0
HiL-Bench (Human-in-Loop Benchmark)Agents39.765.3+25.6
BrokenArXiv 04/2026Math72.197.5+25.4
TextQuests Game ProgressGames48.974.2+25.3
FrontierMath Tier 4 (v2) - PrivateMath72.597.6+25.1
AutomationBench 1.0.6 (Zapier)Agents16.641.4+24.8
GlobeBenchGeneral QA25.249.6+24.4
RuneBenchGames5.77.3+24.3
ChessBench EloGames13491755+24.2
CadQueryEval (Corrected Geometry)Code76.0100.0+24.0
ZeroBench Main Questions pass^5Multimodal11.035.0+24.0
TxBench Antibody Discovery V0Science33.656.9+23.3
AA-Briefcase EloAgents11361534+23.1
Code MigrationCode45.267.7+22.6
OpenAI MRCR v2 8-needle - 512K-1MGeneral QA74.096.3+22.3
ArXivMath 04/2026Math68.890.8+22.1
BullshitBench v2 LeaderboardSafety47.069.0+22.0
TxBench Oligo Discovery V0Science33.655.5+21.8
AutomationBench-AA 1.0.6Agents47.368.5+21.2
HPCT Refusal (CAIS)Safety29.050.0+21.0
BioSecBench-Function V1 Clean RerunsScience28.348.6+20.3
LMArena Agent Arena Net ImprovementAgents4.210.4+20.2
Harvey's Legal Agent Benchmark - Corporate M&ALegal0.020.0+20.0
Harvey's Legal Agent Benchmark - Data Privacy CybersecurityLegal20.00.0−20.0
Harvey's Legal Agent Benchmark - Energy Natural ResourcesLegal20.00.0−20.0
Harvey's Legal Agent Benchmark - Trusts Estates Private ClientLegal0.020.0+20.0
TaxCalcBench TY25Finance & Accounting24.044.0+20.0
ScBench-Long V0 VerifiedScience22.742.4+19.7
TxBench Antibody Discovery V1 Clean RerunsScience29.948.9+19.0
SpatialViz-BenchMultimodal74.293.1+18.9
VariantBenchScience29.748.6+18.9
FrontierCode 2.0 - Pass RateCode26.445.1+18.8
RedlineBenchLegal44.162.8+18.7
OpenAI MRCR v2 8-needle - 256K-512KGeneral QA81.5100.0+18.5
RSI-Exam 0.1 (Full 88)Agents0.30.5+18.1
FrontierCode 2.0 - ScoreCode23.641.7+18.1
PostTrainBench v1.2Indexes23.941.9+18.0
IntPhys 2Multimodal59.977.4+17.5
ArXivMath 05/2026Math77.595.0+17.5
LiveBench 2.0Agents68.786.1+17.4
Roboflow Vision Evals (October 2026, five tasks)Multimodal66.083.4+17.4
TxBench Oligo Discovery V1 Clean RerunsScience31.949.2+17.2
Medical Long Context Reasoning (MLCR-AA)Health & Medicine17.835.0+17.2
HieroglyphBenchMultimodal22.840.0+17.2
DrugDiscoveryBenchScience51.668.7+17.1
GBENCH One-shot Coding GScoreGames62.279.2+17.0
SnakeBench Average ApplesGames16.519.8+16.9
Creative Writing v3 EloChat & Writing18452177+16.8
RiemannBenchMath55.272.0+16.8
DeepsecBench — Revalidated V1Code21.137.8+16.7
DeepsecBench — Revalidated V1 (233 issues)Code21.037.6+16.6
Machiavelli Deception (CAIS)Safety81.776.3+16.3
MMMU-Pro Extended (Mercor held-out)Multimodal43.860.0+16.2
VariantBench V1 Clean RerunsScience30.546.6+16.1
Document ArenaProfessional Work14841467−15.2
SimpleBenchGeneral QA69.083.6+14.6
Artificial Analysis Intelligence Index v4.3.2Indexes38.452.7+14.3
SWE Atlas - RefactoringCode44.859.0+14.3
Artificial Analysis Intelligence Index v4.3Indexes38.652.8+14.2
NoCoT-Bench NCKI (kspine v3)General QA111131+14.1
ErdosBenchMath2.73.2+13.8
SWE Atlas - Codebase QnACode45.459.1+13.7
Blueprint-Bench 2Multimodal0.40.5+13.5
SUPERChemScience66.880.3+13.5
BuseyBench SVGMultimodal6.07.3+13.2
MindCube TinyMultimodal79.992.7+12.8
VCT v2 (SecureBio)Science45.358.0+12.7
SimpleQA Verified (Epoch anti-abstention)General QA63.075.6+12.6
BioSecBench-Function V0Science36.548.6+12.1
APEX-SWECode36.648.6+12.0
ERQAMultimodal70.582.3+11.8
Roboflow Vision Evals (September 2026 protocol)Multimodal74.886.6+11.8
AA-Omniscience IndexIndexes20.543.7+11.6
LLM Creative Story WritingChat & Writing2.43.4+11.4
LAB-Bench ProtocolQA (SecureBio)Science77.989.3+11.4
Vals RSI Index v1.1 (historical five tasks)Agents17.929.0+11.1
LLM Debate BenchmarkChat & Writing15281606+11.1
Chess Puzzles (task version 2.0)Games63.074.0+11.0
HealthBench ProfessionalHealth & Medicine57.268.2+11.0
ArXivMath 06/2026Math83.694.4+10.8
VulcanBench Frontier v4Code78.589.2+10.7
Vals RSI Index v1.1 (four tasks)Agents16.527.1+10.6
Artificial Analysis Progress Index v1Indexes320350+10.4
LMCAGeneral QA54.364.4+10.1
MedXpertQA MM (200-question set)Health & Medicine76.786.8+10.1
ARC-AGI-2General QA85.095.0+10.0
SnakeBench Win RateGames90.080.0−10.0
Humanity's Last ExamGeneral QA43.853.6+9.8
APEX-AccountingFinance & Accounting48.257.9+9.7
GDP.pdf (Artificial Analysis, All-pass)Professional Work22.632.2+9.6
GDPval Extended (Mercor held-out rubric score)Professional Work54.564.1+9.6
GDPval-AA v2.1Professional Work13571572+9.5
Vals Index v2Indexes57.466.6+9.2
Conceptual Reasoning IndexIndexes67.373.9+9.0
DeepSWE v1.1Code64.473.2+8.8
EnterpriseBench: CoreCraftProfessional Work51.342.6−8.7
HardGeoBench Avg DistanceMultimodal11961478−8.7
DAXBench v2Code88.396.8+8.5
FrontierMath Tiers 1-3 (v2) - PrivateMath85.393.7+8.4
MASKSafety90.198.5+8.4
Finance Agent v2 - PrecedentsFinance & Accounting21.930.3+8.4
WeirdML (v2)Science84.993.3+8.4
ComplexConstraintsChat & Writing49.557.7+8.2
SWE Atlas - Test WritingCode42.650.7+8.2
GDPval-AA v2Professional Work13721530+7.8
LisanBench Path LengthGames98457393−7.6
HardGeoBench <=10miMultimodal28.836.3+7.5
BioSecBench-RefusalSafety37.144.6+7.5
HardGeoBench Country AccuracyMultimodal45.052.5+7.5
Harvey HLAB — Criteria Pass RateLegal80.087.5+7.5
Epoch Capabilities Index (ECI)Indexes160169+7.4
SciCode (main problems)Science12.319.6+7.3
Senior SWE-BenchCode29.536.8+7.3
Excel Modeling BenchmarkFinance & Accounting64.571.7+7.2
SpatialBench-Long V1 Clean RerunsScience32.139.3+7.1
Agents' Last Exam (ALE)Agents24.531.6+7.1
BrowseCompSearch84.491.5+7.1
ScBench V1 Clean RerunsScience57.163.9+6.8
MBCT (SecureBio)Science57.163.9+6.8
OncologyBench Strict (25 tasks, October 9 rubric)Health & Medicine14.721.3+6.7
ChessBench AccuracyGames78.885.0+6.6
BioTIER RefusalSafety73.479.9+6.5
CritPt (avg@5)Science25.431.7+6.3
HardGeoBench <=100miMultimodal40.046.3+6.3
VCT (SecureBio)Science52.358.5+6.2
Finance Agent v2 - ComparablesFinance & Accounting43.737.8−5.9
ScBench-Long V1 Clean RerunsScience21.627.3+5.7
FrontierCode 1.1 Extended - ScoreCode56.762.1+5.5
Finance Agent v2 - Market AnalysisFinance & Accounting60.365.5+5.3
MMMU ProMultimodal81.286.4+5.3
FrontierCode 1.1 Extended - Pass RateCode62.868.0+5.2
SAGEMath51.546.4−5.2
GPQA DiamondGeneral QA90.795.8+5.1
HealthBench Hard (length-adjusted)Health & Medicine31.536.6+5.1
Finance Agent v2 - General Quantitative AnalysisFinance & Accounting73.778.8+5.1
ProgramBenchCode0.55.5+5.0
LAB-Bench LitQA2 (SecureBio)Science49.854.8+5.0
SpatialBench V0 VerifiedScience65.270.1+4.9
Terminal-Bench 2.1Code84.389.1+4.9
Bio Capabilities Index (SecureBio)Indexes158164+4.8
APEX-Agents 1.1Professional Work70.375.1+4.8
DeepResearch Bench II (Mercor 130-task set)Search45.450.1+4.7
Longform Creative WritingChat & Writing78.282.8+4.6
SM BenchSafety65.469.9+4.5
AA-LCR v1.1Search84.380.0−4.3
LMArena Agent Arena SteerabilityAgents6.9-1.4−4.1
𝜏³-BankingAgents39.043.1+4.1
TxBench Preclinical Pharmacology V1 Clean RerunsScience55.359.3+4.0
Finance Agent v2 - Earnings AnalysisFinance & Accounting73.269.5−3.7
ARC-AGI-1 Semi-PrivateGeneral QA95.098.5+3.5
SpatialBench V1 Clean RerunsScience64.167.5+3.5
LLM Chess EloGames15321614+3.4
ChessBench CoherenceGames100.097.0−3.0
CRED v0.92Science96.599.5+3.0
SnakeBench Best ApplesGames26.027.0+2.9
FrontierMath ErdősMath0.02.9+2.9
WCB (SecureBio)Science59.762.6+2.9
Tax Agent BenchFinance & Accounting60.563.3+2.9
Finance Agent v2 - General Qualitative AnalysisFinance & Accounting65.668.4+2.9
ITBench-AAAgents45.848.6+2.8
AHK-EvalCode97.2100.0+2.8
Finance Agent v2 - Disclosure AnalysisFinance & Accounting58.961.7+2.8
LAB-Bench CloningScenarios (SecureBio)Science97.3100.0+2.7
PCT Political Inconsistency (CAIS)Safety40.037.4+2.6
MMLU ProGeneral QA88.991.5+2.6
Finance Agent v2 - AdjustmentsFinance & Accounting45.748.3+2.6
Ruby LLM Benchmarks — Program FixerCode77.675.1−2.5
CharXiv Extended (Mercor held-out)Multimodal82.785.1+2.4
HPCT (SecureBio)Science66.869.0+2.2
LisanBench Difficulty-WeightedGames33163113−2.2
GDPval (Mercor 213-task rubric score)Professional Work84.686.7+2.1
matplotgymCode55.958.0+2.1
LMArena Agent Arena Bash RecoveryAgents9.55.5−2.0
Extended NYT Connections 940 (Quadratic v1)Games96.298.1+1.9
HealthBench (length-adjusted)Health & Medicine56.558.3+1.8
SpeciEval (Pooled English Runs)Safety98.296.5−1.7
SpeciEval (Refusal-Aware Retries, Historical)Safety98.296.5−1.7
VCT-Refusal (CAIS)Safety16.017.6+1.6
SpatialBench-Long V0 VerifiedScience33.334.9+1.5
HealthBench (updated implementation)Health & Medicine58.456.9−1.5
CharXiv (descriptive and reasoning combined)Multimodal94.696.1+1.5
Toloka Arena (v1)Agents52.353.8+1.5
BioSecBench-Surveillance V1 Clean RerunsScience47.148.5+1.4
SnakeBench RatingGames37.437.9+1.4
Finance Agent v2 - Financial ModelingFinance & Accounting22.921.6−1.3
DTBenchGeneral QA96.097.3+1.3
SnakeBench Total ApplesGames181198+1.3
LMArena Text Arena EloChat & Writing14811474−1.3
AA-AnalystAgent (pass^5)Professional Work50.051.3+1.2
ACCoRDGeneral QA80.581.7+1.2
MedScribeHealth & Medicine86.987.9+1.0
Legal Research BenchLegal40.439.4−1.0
APEX-1 (Mercor, mean score)Professional Work68.469.3+0.9
AdvancedIFChat & Writing81.382.0+0.7
PubMedQA (SecureBio)Health & Medicine80.581.2+0.7
Finance Agent v2 - All PassFinance & Accounting39.640.3+0.7
MedCodeHealth & Medicine49.148.5−0.6
Vectara Hallucination LeaderboardSafety9.38.7+0.6
SEA-HELMGeneral QA76.376.9+0.6
HealthBench Hard (updated implementation)Health & Medicine33.834.2+0.4
ForecastBench Preliminary Dataset Brier IndexForecasting63.362.9−0.4
SciCode 1.0.1 (Artificial Analysis)Code56.156.5+0.3
EpiBenchScience45.044.6−0.3
HealthBench Consensus (updated implementation)Health & Medicine95.795.4−0.3
Creative Writing v3 Rubric ScoreChat & Writing15.915.9+0.2
ForecastBench Tournament Overall Brier IndexForecasting67.066.9−0.1
Vision Arena OverallMultimodal128312830.0
AIME 2025Math100.0100.00.0
BabaIsBench Stage 0: The IntroGames100.0100.00.0
EpiBench V1 Clean RerunsScience28.328.30.0
GPQA-Bio (SecureBio)Science87.487.40.0
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 - Banking FinanceLegal0.00.00.0
Harvey's Legal Agent Benchmark - Bankruptcy RestructuringLegal0.00.00.0
Harvey's Legal Agent Benchmark - Capital MarketsLegal0.00.00.0
Harvey's Legal Agent Benchmark - Corporate GovernanceLegal0.00.00.0
Harvey's Legal Agent Benchmark - Emerging Companies Venture CapitalLegal0.00.00.0
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 - Healthcare Life SciencesLegal0.00.00.0
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 - Structured Finance SecuritizationLegal0.00.00.0
Harvey's Legal Agent Benchmark - TaxLegal0.00.00.0
Harvey's Legal Agent Benchmark - White Collar Defense InvestigationsLegal0.00.00.0
MedXpertQA MM Extended (Mercor held-out)Health & Medicine60.860.80.0
ObviousBenchGeneral QA100.0100.00.0
OTIS Mock AIME 2024-2025Math100.0100.00.0

GPT-5.5vsGPT-6 Astranormalised score marginsData Nov 15, 2026Data: Vals AI, benchmarks.bio, Artificial Analysis and 67 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…