Atlas

Compare Models

Claude Haiku 5.5 vs Gemini 4 Hydrogen

Gemini 4 Hydrogen leads by 4.5 capability points (≈95% interval +2.1 to +6.9): a clear lead over Claude Haiku 5.5 · P(higher) ≈ 100%

Wins 58 of 64 shared benchmarks (6 lost)

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

Benchmark wins6 – 58 · 64 shared

Comparing matching setups only (2 benchmarks excluded) · suites counted once
← Claude Haiku 5.5Gemini 4 Hydrogen →
Data: Artificial Analysis, Vals AI, Epoch AI and 16 othersatlas.village.dev

Margins

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

SimpleQA Verified (Epoch anti-abstention)Knowledge & reasoning › General knowledge23.870.9+47.1
FrontierMath Tier 4 (v2) - PrivateMath46.387.8+41.5
TaxCalcBench TY25Professional › Finance, accounting & tax4.044.0+40.0
AutomationBench 1.0.6 (Zapier)Tool & computer use › Customer service & operations12.051.1+39.1
AutomationBench-AA 1.0.6Tool & computer use › Customer service & operations36.074.4+38.5
Mystery Game PuzzlesGames30.065.0+35.0
SimpleBenchKnowledge & reasoning › General knowledge59.291.7+32.5
Code Migration v1.1 (Vals Index v2.2 subset)Software › Software engineering50.282.7+32.5
Terminal-Bench-Science 0.1Science › Multidisciplinary18.149.5+31.4
ARC-AGI-2Knowledge & reasoning › Abstract reasoning61.392.5+31.2
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge29.658.8+29.2
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents-0.613.5+27.5
EBR-bench v4 (card ban, single agent)Games43.870.5+26.7
Terminal-Bench 4.0 (Artificial Analysis)Software › Software engineering29.353.5+24.2
BuseyBench SVGPerception & media › Generation4.66.9+23.0
Artificial Analysis Strategy & Ops IndexProfessional › General knowledge work39.157.7+18.6
Artificial Analysis Legal IndexProfessional › Legal38.756.7+18.1
Terminal-Bench 4.0Software › Software engineering36.454.0+17.7
Artificial Analysis Engineering IndexScience › Engineering42.560.1+17.6
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents-5.529.3+17.4
AA-Omniscience v1.1 IndexKnowledge & reasoning › General knowledge12.346.3+17.0
Tax Agent BenchProfessional › Finance, accounting & tax62.278.4+16.2
Artificial Analysis Healthcare & Medical IndexHealth & medicine37.553.3+15.8
Artificial Analysis Finance & Accounting IndexProfessional › Finance, accounting & tax43.658.1+14.5
CritPt v2 (agentic, avg@5)Science › Physical sciences66.380.6+14.3
LMArena Code Arena WebDevSoftware › Frontend & design15961703+14.3
Ramp SWE-Bench v2Software › Software engineering57.471.6+14.2
CursorBench 4.1Software › Software engineering48.461.4+13.0
FrontierMath Tiers 1-3 (v2) - PrivateMath75.188.1+13.0
MultiChallengeLanguage & writing › Conversation67.780.6+12.9
Artificial Analysis Economics IndexComposite indexes42.755.0+12.4
GDP.pdf (Artificial Analysis, All-pass)Professional › General knowledge work20.832.8+12.0
LMArena Text Arena EloLanguage & writing › General assistance14211490+11.9
GBENCH One-shot Coding GScoreGames55.466.8+11.5
FrontierCode 2.0 - Pass RateSoftware › Software engineering31.943.2+11.3
Bug Hunt BenchSoftware › Software engineering20.531.1+10.6
Excel Modeling Benchmark v1.1Professional › Finance, accounting & tax65.575.8+10.3
Legal Research BenchProfessional › Legal43.853.8+10.1
SWE-Marathon v1.2Software › Software engineering48.158.1+10.0
Furniture Assembly BenchmarkPerception & media › Spatial & visual reasoning47.557.5+10.0
FrontierCode 2.0 - ScoreSoftware › Software engineering28.938.9+9.9
Vals Index v2.2Composite indexes58.768.5+9.8
ARC-AGI-1 Semi-PrivateKnowledge & reasoning › Abstract reasoning89.098.5+9.5
FrontierSWE V2Software › Software engineering43.853.1+9.3
Vals Finance IndexProfessional › Finance, accounting & tax57.466.1+8.7
LMArena Agent Arena Steering BurdenTool & computer use › General agents0.817.4+8.3
LiveBench 2.0 (2026-11-06 release)Composite indexes68.977.0+8.1
FORTRESS Average Risk Score (ARS)Safety & alignment9.517.5−8.0
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine35.027.2−7.8
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents0.215.7+7.7
HLAB-Vals-1.1Professional › Legal4.211.3+7.1
DualEntry Accounting AI BenchmarkProfessional › Finance, accounting & tax76.283.2+7.0
CursorBench 4.0Software › Software engineering48.454.3+5.9
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents6.6-4.5−5.5
Epoch Capabilities Index (ECI)Composite indexes160167+5.2
Chess Puzzles (task version 2.0)Games55.060.0+5.0
Artificial Analysis Intelligence Index v5.0.1Composite indexes36.541.2+4.7
Vals Legal IndexProfessional › Legal70.373.6+3.3
Vibe Code Bench v1.1Software › Frontend & design90.487.8−2.7
AA-Briefcase v1.1 EloProfessional › General knowledge work15231569+2.5
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling-1.50.7+1.1
OTIS Mock AIME 2024-2025Math98.9100.0+1.1
SWE-Marathon v1.1Software › Software engineering50.049.4−0.6
Finance Agent v2.1Professional › Finance, accounting & tax65.465.1−0.3
Data: Artificial Analysis, Vals AI, Epoch AI and 16 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…