Atlas

Compare Models

Claude Sonnet 5.5 vs Gemini 4 Hydrogen

Claude Sonnet 5.5 leads by 0.9 capability points (≈95% interval −1.8 to +3.5): statistically tied · P(higher) ≈ 74%

Gemini 4 Hydrogen wins 40 of 76 shared benchmarks (32 lost, 4 tied)

Capability vs. relative cost
Claude Sonnet 5.5Gemini 4 HydrogenClosed weightsOpen weights

Benchmark wins32 – 40 · 76 shared, 4 tied

Comparing matching setups only (2 benchmarks excluded) · suites counted once
← Claude Sonnet 5.5Gemini 4 Hydrogen →
Data: Artificial Analysis, Surge AI, Epoch AI and 24 othersatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Gemini 4 Hydrogen minus Claude Sonnet 5.5.

TaxCalcBench TY25Professional › Finance, accounting & tax18.044.0+26.0
SimpleQA Verified (Epoch anti-abstention)Knowledge & reasoning › General knowledge46.570.9+24.4
TxBench-Antibody-Development V1 (60 evaluations)Science › Life sciences36.760.6+23.9
SWE-Marathon v1.2Software › Software engineering80.658.1−22.5
sudo L7 (October 2026 rescoring)Software › Software engineering50.631.7−18.9
Bug Hunt BenchSoftware › Software engineering48.931.1−17.8
Furniture Assembly BenchmarkPerception & media › Spatial & visual reasoning75.057.5−17.5
DAYJOB: Law (November 2026 release)Professional › Legal26.89.7−17.1
SimpleBenchKnowledge & reasoning › General knowledge75.991.7+15.8
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine41.727.2−14.4
Artificial Analysis Intelligence Index v5.0.1Composite indexes54.541.2−13.3
GDP.xlsxProfessional › General knowledge work29.140.9+11.8
Agents' Last Exam v1.1 ScoreProfessional › General knowledge work60.471.6+11.2
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine34.024.0−10.0
AA-Briefcase v1.1 EloProfessional › General knowledge work17391569−9.3
LMArena Code Arena WebDevSoftware › Frontend & design17731703−9.3
FrontierSWE V2Software › Software engineering61.953.1−8.8
ARC-AGI-2Knowledge & reasoning › Abstract reasoning83.992.5+8.6
Ramp SWE-Bench v2Software › Software engineering63.271.6+8.4
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents11.9-4.5−8.2
GradeBenchProfessional › Education43.751.9+8.2
BigFinanceBench private holdout (final-answer accuracy)Professional › Finance, accounting & tax41.549.4+7.9
AA-Omniscience v1.1 IndexKnowledge & reasoning › General knowledge30.546.3+7.9
FrontierMath Tier 4 (v2) - PrivateMath80.587.8+7.3
DAYJOB: TaxProfessional › Finance, accounting & tax10.83.6−7.2
GDP.pdf (Artificial Analysis, All-pass)Professional › General knowledge work25.832.8+7.0
CursorBench 4.1Software › Software engineering54.461.4+7.0
CWE-bench v2 Pass@1Software › Cybersecurity61.468.1+6.7
ComplexConstraints (November 2026 rescoring)Language & writing › Instruction following67.273.7+6.5
SpireClimbGames1.37.7+6.3
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge53.358.8+5.5
CritPt v2 (agentic, avg@5)Science › Physical sciences75.180.6+5.4
ChartographyPerception & media › Documents & charts61.666.6+5.0
Vibe Code Bench v1.1Software › Frontend & design92.487.8−4.6
Artificial Analysis Legal IndexProfessional › Legal52.256.7+4.6
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents15.613.5−4.0
Terminal-Bench 4.0 (Artificial Analysis)Software › Software engineering57.153.5−3.5
Gray Swan IPI Attack Success (k=15)Safety & alignment3.40.0+3.4
FrontierCode 2.0 - ScoreSoftware › Software engineering42.238.9−3.4
BuseyBench SVGPerception & media › Generation6.56.9+3.3
FrontierCode 2.0 - Pass RateSoftware › Software engineering46.443.2−3.2
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations47.744.6−3.1
TxBench Antibody Discovery V1 (100 evaluations)Science › Life sciences54.757.7+3.0
DualEntry Accounting AI BenchmarkProfessional › Finance, accounting & tax80.283.2+3.0
GDP.pdf (November 2026 rescoring)Professional › General knowledge work51.254.2+3.0
TxBench Preclinical Pharmacology V1 Clean RerunsScience › Life sciences64.361.3−3.0
Terminal-Bench-Science 0.1Science › Multidisciplinary52.449.5−2.9
LMArena Text Arena EloLanguage & writing › General assistance14741490+2.8
AutomationBench-AA 1.0.6Tool & computer use › Customer service & operations71.874.4+2.7
TxBench Oligo Discovery V1 (113 evaluations)Science › Life sciences51.954.3+2.4
Artificial Analysis Finance & Accounting IndexProfessional › Finance, accounting & tax55.858.1+2.3
ARC-AGI-1 Semi-PrivateKnowledge & reasoning › Abstract reasoning96.598.5+2.0
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax40.538.5−2.0
SWE-Marathon v1.1Software › Software engineering51.349.4−1.9
Artificial Analysis Engineering IndexScience › Engineering58.560.1+1.6
DAYJOB: AccountingProfessional › Finance, accounting & tax35.233.6−1.6
LMArena Agent Arena Steering BurdenTool & computer use › General agents14.817.4+1.3
CursorBench 4.0Software › Software engineering55.554.3−1.2
Chess Puzzles (task version 2.0)Games59.060.0+1.0
Artificial Analysis Economics IndexComposite indexes54.155.0+1.0
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents17.615.7−0.9
Artificial Analysis Healthcare & Medical IndexHealth & medicine54.253.3−0.9
HLAB-Vals-1.1Professional › Legal10.411.3+0.8
FORTRESS Average Risk Score (ARS)Safety & alignment18.217.5+0.7
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents30.729.3−0.7
FrontierMath Tiers 1-3 (v2) - PrivateMath88.888.1−0.7
GBENCH One-shot Coding GScoreGames66.466.8+0.4
Artificial Analysis Strategy & Ops IndexProfessional › General knowledge work58.157.7−0.3
LiveBench 2.0 (2026-11-06 release)Composite indexes76.877.0+0.2
Epoch Capabilities Index (ECI)Composite indexes167167+0.2
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling1.00.7−0.1
Excel Modeling Benchmark v1.1Professional › Finance, accounting & tax75.975.8−0.1
Dual-Arm Sim (task completion)Tool & computer use › Robotics0.00.00.0
micro1 PharmaBenchProfessional › General knowledge work11.411.40.0
Mystery Game PuzzlesGames65.065.00.0
OTIS Mock AIME 2024-2025Math100.0100.00.0
Data: Artificial Analysis, Surge AI, Epoch AI and 24 othersatlas.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…