Atlas

Compare Models

Opus 4.8 vs Claude Opus 5.5

Claude Opus 5.5 leads by 8.0 capability points (≈95% interval +5.7 to +10.3): a clear lead over Opus 4.8 · P(higher) ≈ 100%

Wins 172 of 214 shared benchmarks (21 lost, 21 tied)

Capability vs. relative cost
Opus 4.8Claude Opus 5.5Closed 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.

Opus 4.8vsClaude Opus 5.5capability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins21 – 172 · 214 shared, 21 tied

Comparing matching setups only (27 benchmarks excluded) · suites counted once
← Opus 4.8Claude Opus 5.5 →

Opus 4.8vsClaude Opus 5.5wins by categoryData Nov 15, 2026Data: Vals AI, Mercor, Center for AI Safety and 65 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Claude Opus 5.5 minus Opus 4.8.

Time Horizon Index: KSPAgents11.891.3+79.5
BrokenArXiv 05/2026Math23.095.0+72.0
VoxelBench Text RatingGames16272613+70.2
Concept Synth INDUCTION Challenge64 Round 1General QA6.371.9+65.6
TasteVal (compute multiplier)Science0.82.3+64.2
KernelBench Mega Normalized (RTX PRO 6000)Code40.6100.0+59.4
BrokenArXiv 06/2026Math41.7100.0+58.3
KernelBench CUDA Normalized (RTX PRO 6000)Code24.976.3+51.3
Clocktower BenchGames11951547+50.5
VCT-Refusal (CAIS)Safety41.290.4+49.2
LLM Chess EloGames4501575+46.2
LisanBench Path LengthGames946324029+45.3
Terminal-Bench 4.0Code23.664.8+41.2
Furniture Assembly BenchmarkMultimodal42.583.3+40.8
YC-BenchGames17113892873725+40.6
FrontierMath Tier 4 (v2) - PrivateMath56.195.0+38.9
ActiveVisionMultimodal2.441.2+38.8
Terminal-Bench Science 0.1 (Vals)Science10.048.6+38.6
Terminal-Bench 4.0 (Artificial Analysis)Code21.759.6+37.9
EnigmaEvalGeneral QA20.557.8+37.3
Blueprint-Bench 2Multimodal0.10.5+36.7
LisanBench Difficulty-WeightedGames26945963+35.0
Mystery Game PuzzlesGames36.071.0+35.0
FrontierCode 2.0 - Pass RateCode17.151.8+34.7
TxBench Oligo Discovery V1 Clean RerunsScience35.31.4−33.9
LMArena Code Arena WebDevCode15551814+33.9
LLM Creative Story WritingChat & Writing0.83.7+33.4
GameBench 2Games62.594.8+32.3
FrontierCode 2.0 - ScoreCode15.347.3+32.0
CadQueryEval (Corrected Geometry)Code68.0100.0+32.0
Humanity's Last Exam DiamondGeneral QA23.154.6+31.5
BullshitBench v2 LeaderboardSafety94.063.0−31.0
BioSecBench-Function V1 Clean RerunsScience43.173.6+30.5
BrowseComp Extended (Mercor held-out)Search30.060.5+30.5
MindCube TinyMultimodal64.993.1+28.2
AURORA Multilingual Terminal-BenchCode49.176.5+27.3
AA-Briefcase v1.1 EloAgents13201801+26.3
TxBench Antibody Discovery V1 Clean RerunsScience32.46.1−26.3
ZeroBench (Main questions, pass@5)Multimodal17.043.0+26.0
Bug Hunt BenchCode14.339.7+25.4
Into the Unknown Integrity Node Pass RateScience65.390.4+25.1
HardGeoBench <=100miMultimodal36.361.3+25.0
PCT Political Inconsistency (CAIS)Safety53.829.1+24.7
SpatialViz-BenchMultimodal65.289.7+24.5
RuneBenchGames5.16.7+24.2
Roboflow Vision Evals (October 2026, five tasks)Multimodal58.382.0+23.7
APEX-SWECode43.967.6+23.7
HiL-Bench (Human-in-Loop Benchmark)Agents35.359.0+23.7
TxBench Preclinical Pharmacology V1 Clean RerunsScience54.330.7−23.7
SimpleBenchGeneral QA64.888.4+23.6
HieroglyphBenchMultimodal14.037.4+23.4
ChessBench EloGames9611349+23.1
ALE-BenchCode21022748+22.9
Agent Red Teaming k=100 (CAIS)Safety47.024.4+22.6
LAB-Bench LitQA2 (SecureBio)Science48.771.1+22.4
RiemannBenchMath47.269.6+22.4
EnterpriseBench: CoreCraftProfessional Work52.374.4+22.1
ReactBench (39 tasks)Code32.354.4+22.1
TaxCalcBench TY25Finance & Accounting20.042.0+22.0
Design Arena Overall Code CategoriesCode12521384+21.9
OncologyBench Strict (25 tasks, October 9 rubric)Health & Medicine16.037.3+21.3
ARC-AGI-2General QA72.193.3+21.3
Medical Long Context Reasoning (MLCR-AA)Health & Medicine45.666.7+21.1
Humanity's Sixth SenseMultimodal23.544.6+21.1
MineBench — Bradley-Terry RatingGames17862134+21.1
ExploitGymCybersecurity13.834.5+20.7
Harvey's Legal Agent Benchmark - Capital MarketsLegal20.00.0−20.0
Harvey's Legal Agent Benchmark - Data Privacy CybersecurityLegal20.00.0−20.0
Harvey's Legal Agent Benchmark - Emerging Companies Venture CapitalLegal20.00.0−20.0
Harvey's Legal Agent Benchmark - Healthcare Life SciencesLegal40.020.0−20.0
Harvey's Legal Agent Benchmark - Structured Finance SecuritizationLegal20.00.0−20.0
GBENCH One-shot Coding GScoreGames54.073.9+19.9
Into the Unknown Direction Node Pass RateScience60.180.0+19.8
DeepSWE v1.1Code54.474.2+19.8
DeepsecBench — Revalidated V1Code6.926.7+19.8
DeepsecBench — Revalidated V1 (233 issues)Code6.926.6+19.7
MedXpertQA MM Extended (Mercor held-out)Health & Medicine51.671.2+19.6
MMMU-Pro Extended (Mercor held-out)Multimodal31.451.0+19.6
SUPERChemScience60.379.6+19.3
SWE Atlas Codebase QnA Extended (Mercor held-out)Code24.043.3+19.3
SimpleQA Verified (Epoch anti-abstention)General QA53.072.2+19.2
Machiavelli Deception (CAIS)Safety83.777.5+18.9
GDPval-AA v2.1Professional Work14611866+17.8
HardGeoBench Avg DistanceMultimodal1245669+17.8
TextQuests Game ProgressGames40.157.3+17.2
LAB-Bench ProtocolQA (SecureBio)Science76.994.0+17.1
GDPval Extended (Mercor held-out rubric score)Professional Work58.975.8+16.9
APEX-Agents 1.1Professional Work64.481.3+16.9
Roboflow Vision Evals (September 2026 protocol)Multimodal68.785.5+16.8
ArXivMath 05/2026Math65.081.7+16.7
LVBenchMultimodal67.383.7+16.4
BuseyBench SVGMultimodal5.26.8+16.4
Agents' Last Exam (ALE)Agents22.438.2+15.8
ERQAMultimodal59.575.1+15.6
Drone-BenchAgents82.397.3+15.0
Chess Puzzles (task version 2.0)Games52.067.0+15.0
HPCT Refusal (CAIS)Safety85.0100.0+15.0
HardGeoBench <=10miMultimodal27.542.5+15.0
BrowseComp (Mercor 130-question subset)Search73.788.5+14.8
Terminal-Bench 2.1 Extended (Mercor held-out)Code32.747.5+14.8
VCT v2 (SecureBio)Science45.960.5+14.6
ComplexConstraintsChat & Writing35.649.9+14.3
Chartography (with tools)Multimodal75.089.0+14.0
Vals RSI Index v1.1 (historical five tasks)Agents23.237.1+13.9
APEX-AccountingFinance & Accounting48.061.8+13.8
HardGeoBench Country AccuracyMultimodal53.867.5+13.8
ArXivMath 06/2026Math70.083.3+13.4
Vals RSI Index v1.1 (four tasks)Agents24.037.3+13.3
MedXpertQA MM (200-question set)Health & Medicine72.585.7+13.2
EpiBench V1 Clean RerunsScience25.839.0+13.2
VCT (SecureBio)Science48.261.4+13.2
Conceptual Reasoning IndexIndexes68.377.9+13.2
IntPhys 2Multimodal55.168.3+13.2
SpatialBench-Long V1 Clean RerunsScience33.320.2−13.1
Olam Social Poker EloGames15181506−13.0
Remote Labor Index (RLI)Professional Work8.321.3+12.9
PostTrainBench v1.2Indexes31.043.8+12.8
Humanity's Last Exam (Text-Only)General QA44.857.5+12.7
Humanity's Last Exam RMS Calibration Error (CAIS)General QA26.413.8+12.6
SciCode 1.0.1 (Artificial Analysis)Code54.466.9+12.5
MBCT (SecureBio)Science56.468.7+12.3
WCB (SecureBio)Science66.754.5−12.2
Harvey HLAB — Criteria Pass RateLegal87.975.8−12.1
GeneBench-Pro Public Release SubsetScience18.030.0+12.0
PubMedQA (SecureBio)Health & Medicine75.987.9+12.0
FACTS Benchmark SuiteGeneral QA53.565.3+11.8
AA-AnalystAgent (pass^5)Professional Work45.056.3+11.3
FrontierMath Tiers 1-3 (v2) - PrivateMath80.091.2+11.2
HealthBench ProfessionalHealth & Medicine60.371.4+11.1
AHK-EvalCode80.691.7+11.1
ACCoRDGeneral QA77.488.4+11.0
AA-LCR Extended (Mercor held-out)Search63.274.2+11.0
TextQuests Average Cumulative HarmSafety21.024.2−10.9
LMCAGeneral QA57.568.2+10.7
SciCode (main problems)Science16.927.3+10.4
Humanity's Last Exam (with tools)General QA57.667.7+10.1
GPQA-Bio (SecureBio)Science80.990.9+10.0
LLM Debate BenchmarkChat & Writing16311701+10.0
LiveBench 2.0Agents76.886.5+9.8
AdvancedIFChat & Writing73.782.8+9.1
SAGEMath54.845.8−9.0
BioSecBench-Surveillance V1 Clean RerunsScience46.955.8+8.9
AA-Omniscience IndexIndexes28.846.4+8.8
SM BenchSafety72.981.8+8.8
Senior SWE-BenchCode30.539.3+8.8
DeepResearch Bench II (Mercor 130-task set)Search44.653.3+8.7
Arena Alignment IndexSafety74.583.2+8.6
VariantBench V1 Clean RerunsScience42.450.6+8.2
CharXiv Extended (Mercor held-out)Multimodal76.784.7+8.0
ChessBench AccuracyGames72.479.3+7.4
ScBench-Long V1 Clean RerunsScience21.013.6−7.4
SWE-TogetherCode61.568.8+7.3
Epoch Capabilities Index (ECI)Indexes160169+7.1
Gray Swan IPI Attack Success (k=15)Safety8.01.0+7.0
SpatialBench V1 Clean RerunsScience65.872.8+7.0
Excel Modeling BenchmarkFinance & Accounting69.475.9+6.6
MetagenomicsBench V1Science50.043.7−6.3
GlobeBenchGeneral QA20.026.3+6.3
MetagenomicsBench V0Science50.043.7−6.3
GDP.pdf (Artificial Analysis, All-pass)Professional Work22.828.8+6.0
ARC-AGI-1 Semi-PrivateGeneral QA92.598.5+6.0
MedScribeHealth & Medicine85.891.4+5.7
HealthBenchHealth & Medicine58.864.3+5.5
Bio Capabilities Index (SecureBio)Indexes164171+5.1
MASKSafety94.489.4−5.0
Ruby LLM Benchmarks — Program FixerCode72.677.5+4.9
SWE Atlas - Codebase QnACode44.448.9+4.5
LMArena Text Arena EloChat & Writing14821506+4.2
DistressBenchSafety71.475.6+4.1
BioTIER RefusalSafety94.798.7+4.0
DTBenchGeneral QA94.998.9+4.0
FrontierCode 1.1 Extended - ScoreCode59.663.6+4.0
FrontierCode 1.1 Extended - Pass RateCode65.569.5+4.0
LMArena Agent Arena Bash RecoveryAgents7.215.0+3.9
GDPval (Mercor 213-task rubric score)Professional Work84.588.3+3.8
WMDP-Bio (SecureBio)Science92.396.1+3.8
MedCodeHealth & Medicine53.249.8−3.4
MMLU ProGeneral QA90.093.4+3.4
CharXiv (descriptive and reasoning combined)Multimodal93.296.4+3.2
APEX-1 (Mercor, mean score)Professional Work62.765.8+3.1
SpeciEval (Pooled English Runs)Safety84.387.2+2.9
SpeciEval (Refusal-Aware Retries, Historical)Safety84.387.2+2.9
SWE-bench Verified Extended (Mercor held-out)Code63.366.0+2.7
Extended NYT Connections 940 (Quadratic v1)Games91.188.5−2.6
RedlineBenchLegal39.742.0+2.3
MMLU-Bio (SecureBio)Science90.992.7+1.8
AIME 2025Math98.3100.0+1.7
OTIS Mock AIME 2024-2025Math98.3100.0+1.7
Terminal-Bench 2.1Code82.083.5+1.5
ObviousBenchGeneral QA99.3100.0+0.7
ForecastBench Preliminary Dataset Brier IndexForecasting61.662.2+0.6
ChessBench CoherenceGames79.780.1+0.4
ScBench V1 Clean RerunsScience56.256.4+0.2
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 - Corporate GovernanceLegal0.00.00.0
Harvey's Legal Agent Benchmark - Corporate M&ALegal0.00.00.0
Harvey's Legal Agent Benchmark - Employment LaborLegal0.00.00.0
Harvey's Legal Agent Benchmark - Energy Natural ResourcesLegal20.020.00.0
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 ClientLegal0.00.00.0
Harvey's Legal Agent Benchmark - White Collar Defense InvestigationsLegal0.00.00.0
Into the UnknownScience0.00.00.0
xbench-ScienceQAScience76.476.40.0

Opus 4.8vsClaude Opus 5.5normalised score marginsData Nov 15, 2026Data: Vals AI, Mercor, Center for AI Safety and 65 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…