Atlas

Compare Models

Opus 4.8 vs Claude Sonnet 5.5

Claude Sonnet 5.5 leads by 6.9 capability points (≈95% interval +4.6 to +9.2): a clear lead over Opus 4.8 · P(higher) ≈ 100%

Wins 123 of 143 shared benchmarks (20 lost)

Capability vs. relative cost
Opus 4.8Claude Sonnet 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 Sonnet 5.5capability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins20 – 123 · 143 shared

Comparing matching setups only (9 benchmarks excluded) · suites counted once
← Opus 4.8Claude Sonnet 5.5 →

Opus 4.8vsClaude Sonnet 5.5wins by categoryData Nov 15, 2026Data: Mercor, benchmarks.bio, SecureBio and 46 othersVillage LLM Atlasatlas.village.dev

Margins

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

KernelBench CUDA Normalized (RTX PRO 6000)Code24.987.0+62.1
KernelBench Mega Normalized (RTX PRO 6000)Code40.690.6+50.0
Concept Synth INDUCTION Challenge64 Round 1General QA6.350.0+43.8
Terminal-Bench Science 0.1 (Vals)Science4.347.1+42.9
Terminal-Bench 4.0 (Artificial Analysis)Code21.763.6+41.9
ActiveVisionMultimodal2.441.2+38.8
Terminal-Bench 4.0Code23.661.8+38.2
Bug Hunt BenchCode14.348.9+34.6
BullshitBench v2 LeaderboardSafety94.060.0−34.0
Furniture Assembly BenchmarkMultimodal42.575.0+32.5
DeepsecBench — Revalidated V1 (233 issues)Code6.937.5+30.6
LLM Chess EloGames4501195+30.6
Humanity's Last Exam DiamondGeneral QA23.153.3+30.2
Medical Long Context Reasoning (MLCR-AA)Health & Medicine45.675.0+29.4
FrontierCode 2.0 - Pass RateCode17.146.4+29.2
Mystery Game PuzzlesGames36.065.0+29.0
GameBench 2Games62.591.4+29.0
LMArena Code Arena WebDevCode15551773+28.4
AA-Briefcase v1.1 EloAgents13201816+27.2
FrontierCode 2.0 - ScoreCode15.342.2+26.9
AutomationBench-AA 1.0.6Agents45.671.8+26.2
Blueprint-Bench 2Multimodal0.10.4+24.8
FrontierMath Tier 4 (v2) - PrivateMath56.180.5+24.4
Code MigrationCode47.369.8+22.6
APEX-SWECode43.966.4+22.5
TxBench Antibody Discovery V1 (100 evaluations)Science35.357.3+22.0
GeneBench-Pro Public Release SubsetScience18.039.0+21.0
SWE Atlas Codebase QnA Extended (Mercor held-out)Code24.044.7+20.7
APEX-Agents 1.1Professional Work64.483.7+19.3
Roboflow Vision Evals (October 2026, five tasks)Multimodal58.377.6+19.3
EnterpriseBench: CoreCraftProfessional Work52.370.8+18.5
HardGeoBench Avg DistanceMultimodal1245683+17.3
GDPval-AA v2.1Professional Work14611838+16.6
DeepSWE v1.1Code54.470.5+16.1
Design Arena Overall Code CategoriesCode12521349+16.1
OncologyBench Strict (25 tasks, October 9 rubric)Health & Medicine16.032.0+16.0
ChessBench CoherenceGames79.764.6−15.1
Roboflow Vision Evals (September 2026 protocol)Multimodal68.783.8+15.1
TxBench Oligo Discovery V1 (113 evaluations)Science37.551.9+14.5
Artificial Analysis Intelligence Index v4.3.2Indexes41.856.0+14.2
HPCT (SecureBio)Science63.177.2+14.1
HardGeoBench <=100miMultimodal36.350.0+13.8
LisanBench Path LengthGames946313787+13.4
GBENCH One-shot Coding GScoreGames54.067.4+13.3
BuseyBench SVGMultimodal5.26.5+13.2
BrowseComp (Mercor 130-question subset)Search73.786.6+12.9
MineBench — Bradley-Terry RatingGames17861998+12.8
Terminal-Bench 2.1 Extended (Mercor held-out)Code32.745.5+12.8
ExploitGymCybersecurity13.826.6+12.8
SM BenchSafety72.985.4+12.5
AA-AnalystAgent (pass^5)Professional Work45.057.5+12.5
ALE-BenchCode21022448+12.3
Conceptual Reasoning IndexIndexes68.377.1+12.1
GDPval Extended (Mercor held-out rubric score)Professional Work58.970.9+12.0
ARC-AGI-2General QA72.183.9+11.8
LAB-Bench ProtocolQA (SecureBio)Science76.988.3+11.4
Artificial Analysis Progress Index v1Indexes318351+11.3
SimpleBenchGeneral QA64.875.9+11.1
ScBench-Long V1 Clean RerunsScience21.031.8+10.8
LMCAGeneral QA57.568.3+10.8
AA-LCR Extended (Mercor held-out)Search63.273.8+10.6
Extended NYT Connections 940 (Quadratic v1)Games91.180.5−10.6
HealthBench ProfessionalHealth & Medicine60.370.8+10.5
MetagenomicsBench V0Science50.060.0+10.0
MetagenomicsBench V1Science50.060.0+10.0
AdvancedIFChat & Writing73.783.4+9.7
Vibe Code Bench v1.1Code82.792.4+9.7
VariantBench V1 Clean RerunsScience42.451.7+9.3
APEX-1 (Mercor, mean score)Professional Work62.753.9−8.8
FrontierMath Tiers 1-3 (v2) - PrivateMath80.088.8+8.8
HardGeoBench <=10miMultimodal27.536.3+8.7
Olam Social Poker EloGames15181526+8.7
BioSecBench-Function V1 Clean RerunsScience43.151.7+8.6
ACCoRDGeneral QA77.485.7+8.3
MedXpertQA MM (200-question set)Health & Medicine72.580.7+8.2
Bio Capabilities Index (SecureBio)Indexes164175+8.1
CharXiv Extended (Mercor held-out)Multimodal76.784.2+7.5
GDP.pdfProfessional Work24.031.4+7.4
DeepResearch Bench II (Mercor 130-task set)Search44.651.7+7.1
LAB-Bench LitQA2 (SecureBio)Science48.755.8+7.1
HealthBenchHealth & Medicine58.865.9+7.1
Chess Puzzles (task version 2.0)Games52.059.0+7.0
NoCoT-Bench NCKI (kspine v3)General QA117107−7.0
SpatialBench V1 Clean RerunsScience65.872.8+7.0
RuneBenchGames5.15.5+6.8
GPQA-Bio (SecureBio)Science80.987.6+6.7
SpeciEval (Pooled English Runs)Safety84.391.0+6.7
Harvey's Legal Agent BenchmarkLegal9.62.9−6.7
SpeciEval (Refusal-Aware Retries, Historical)Safety84.391.0+6.7
TxBench Preclinical Pharmacology V1 Clean RerunsScience57.764.3+6.7
SciCode 1.0.1 (Artificial Analysis)Code54.461.0+6.6
SimpleQA Verified (Epoch anti-abstention)General QA53.046.5−6.5
Excel Modeling BenchmarkFinance & Accounting69.475.7+6.3
VCT v2 (SecureBio)Science45.952.2+6.3
ChessBench EloGames961857−6.2
SWE-bench Verified Extended (Mercor held-out)Code63.357.3−6.0
Vision Arena OverallMultimodal12861263−6.0
SWE Atlas - Codebase QnACode44.450.3+5.9
LiveBench 2.0Agents76.882.5+5.7
ProgramBenchCode1.06.5+5.5
PubMedQA (SecureBio)Health & Medicine75.981.3+5.4
MedScribeHealth & Medicine85.891.1+5.3
Terminal-Bench 2.1Code82.087.3+5.3
Epoch Capabilities Index (ECI)Indexes160166+5.2
HardGeoBench Country AccuracyMultimodal53.858.8+5.0
FrontierCode 1.1 Extended - ScoreCode59.564.4+5.0
FrontierCode 1.1 Extended - Pass RateCode65.069.9+4.8
EpiBench V1 Clean RerunsScience25.830.2+4.4
VCT (SecureBio)Science48.252.5+4.3
MedXpertQA MM Extended (Mercor held-out)Health & Medicine51.655.6+4.0
ARC-AGI-1 Semi-PrivateGeneral QA92.596.5+4.0
LMArena Agent Arena Bash RecoveryAgents7.215.0+3.9
NoCoT-Bench NCRI 15.2General QA10295.1−3.8
Humanity's Last ExamGeneral QA49.846.1−3.7
ChessBench AccuracyGames72.475.8+3.6
ScBench V1 Clean RerunsScience57.160.3+3.2
DistressBenchSafety71.474.6+3.1
LisanBench Difficulty-WeightedGames26942983+3.1
SAGEMath54.851.8−3.0
CharXiv (descriptive and reasoning combined)Multimodal93.296.1+2.9
MBCT (SecureBio)Science56.458.8+2.4
DTBenchGeneral QA94.997.3+2.4
GDPval (Mercor 213-task rubric score)Professional Work84.586.5+2.0
TaxCalcBench TY25Finance & Accounting20.018.0−2.0
AA-Omniscience IndexIndexes28.832.3+1.8
AIME 2025Math98.3100.0+1.7
OTIS Mock AIME 2024-2025Math98.3100.0+1.7
LMArena Text Arena EloChat & Writing14821472−1.6
SpatialBench-Long V1 Clean RerunsScience33.334.9+1.6
Ruby LLM Benchmarks — Program FixerCode72.673.8+1.2
LAB-Bench CloningScenarios (SecureBio)Science95.997.0+1.1
YC-BenchGames17113891741315+1.0
MMLU ProGeneral QA90.091.0+1.0
Public Benefits Bench v1.1Professional Work68.167.2−0.9
SWE-TogetherCode61.562.4+0.9
ObviousBenchGeneral QA99.3100.0+0.7
RedlineBenchLegal39.740.3+0.6
ForecastBench Preliminary Dataset Brier IndexForecasting61.661.1−0.5
ARC-AGI-3 (Semi-Private)General QA1.51.2−0.3
WMDP-Bio (SecureBio)Science92.392.6+0.3
MedCodeHealth & Medicine53.252.9−0.3
WCB (SecureBio)Science66.766.5−0.2
MMLU-Bio (SecureBio)Science90.991.0+0.1

Opus 4.8vsClaude Sonnet 5.5normalised score marginsData Nov 15, 2026Data: Mercor, benchmarks.bio, SecureBio and 46 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…