Atlas

Compare Models

Claude Fable 5.2 vs Gemini 4 Hydrogen

Claude Fable 5.2 leads by 4.1 capability points (≈95% interval +1.2 to +6.9): a clear lead over Gemini 4 Hydrogen · P(higher) ≈ 100%

Wins 50 of 68 shared benchmarks (17 lost, 1 tied)

Capability vs. relative cost
Claude Fable 5.2Gemini 4 HydrogenClosed weightsOpen weights

Benchmark wins50 – 17 · 68 shared, 1 tied

Comparing matching setups only (3 benchmarks excluded) · suites counted once
← Claude Fable 5.2Gemini 4 Hydrogen →
Data: Epoch AI, Surge AI, Arena 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 Fable 5.2.

Furniture Assembly BenchmarkPerception & media › Spatial & visual reasoning90.057.5−32.5
SWE-Marathon v1.2Software › Software engineering86.858.1−28.7
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents27.413.5−27.1
Mystery Game PuzzlesGames90.065.0−25.0
Bug Hunt BenchSoftware › Software engineering55.231.1−24.1
SimpleQA Verified (Epoch anti-abstention)Knowledge & reasoning › General knowledge94.070.9−23.1
HiL-Bench (Human-in-Loop Benchmark)Tool & computer use › General agents77.355.3−22.0
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge80.258.8−21.4
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine45.024.0−21.0
sudo L7 (October 2026 rescoring)Software › Software engineering52.231.7−20.5
EBR-bench v4 (card ban, single agent)Games90.570.5−20.0
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents68.729.3−19.7
Chess Puzzles (task version 2.0)Games79.060.0−19.0
TaxCalcBench TY25Professional › Finance, accounting & tax62.044.0−18.0
FrontierSWE V2Software › Software engineering69.953.1−16.8
DAYJOB: Law (November 2026 release)Professional › Legal26.59.7−16.8
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine43.927.2−16.7
LMArena Code Arena WebDevSoftware › Frontend & design18241703−16.1
FrontierMath Tier 4 (v2) - PrivateMath100.087.8−12.2
DAYJOB: TaxProfessional › Finance, accounting & tax15.63.6−12.0
DAYJOB: AccountingProfessional › Finance, accounting & tax45.233.6−11.6
FrontierCode 2.0 - Pass RateSoftware › Software engineering54.743.2−11.5
Tax Agent BenchProfessional › Finance, accounting & tax67.078.4+11.4
FrontierCode 2.0 - ScoreSoftware › Software engineering50.238.9−11.3
AA-Briefcase v1.1 EloProfessional › General knowledge work17371569−9.2
ComplexConstraints (November 2026 rescoring)Language & writing › Instruction following64.573.7+9.2
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents13.1-4.5−8.8
GBENCH One-shot Coding GScoreGames74.966.8−8.1
GDP.xlsxProfessional › General knowledge work32.940.9+8.0
LiveBench 2.0 (2026-11-06 release)Composite indexes84.977.0−7.9
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations52.344.6−7.7
SpireClimbGames15.37.7−7.7
Dual-Arm Sim (task completion)Tool & computer use › Robotics7.50.0−7.5
FrontierMath Tiers 1-3 (v2) - PrivateMath95.188.1−7.0
CWE-bench v2 Pass@1Software › Cybersecurity75.068.1−6.9
AutomationBench 1.0.6 (Zapier)Tool & computer use › Customer service & operations44.351.1+6.8
ChartographyPerception & media › Documents & charts73.466.6−6.8
HLAB-Vals-1.1Professional › Legal5.011.3+6.3
KradleBenchGames37.843.9+6.1
Vibe Code Bench v1.1Software › Frontend & design93.887.8−6.0
GradeBenchProfessional › Education57.351.9−5.4
Epoch Capabilities Index (ECI)Composite indexes173167−5.3
Professional Reasoning Benchmark - FinanceProfessional › Finance, accounting & tax47.152.3+5.2
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax43.738.5−5.2
Finance Agent v2.1Professional › Finance, accounting & tax69.565.1−4.4
CritPt v2 (agentic, avg@5)Science › Physical sciences84.980.6−4.3
Professional Reasoning Benchmark - LegalProfessional › Legal49.053.1+4.1
Gray Swan IPI Attack Success (k=15)Safety & alignment4.10.0+4.1
Humanity's Sixth SensePerception & media › Spatial & visual reasoning59.563.4+3.9
LMArena Agent Arena Steering BurdenTool & computer use › General agents24.517.4−3.5
Agents' Last Exam v1.1 (ALE v1.1)Professional › General knowledge work46.943.4−3.5
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents22.515.7−3.4
SimpleBenchKnowledge & reasoning › General knowledge94.991.7−3.2
BuseyBench SVGPerception & media › Generation7.26.9−3.1
GPQA DiamondScience › Multidisciplinary93.996.8+2.9
Vals Finance IndexProfessional › Finance, accounting & tax63.666.1+2.6
GDP.pdf (November 2026 rescoring)Professional › General knowledge work51.854.2+2.4
DualEntry Accounting AI BenchmarkProfessional › Finance, accounting & tax81.283.2+2.0
FORTRESS Average Risk Score (ARS)Safety & alignment15.617.5−1.9
Ramp SWE-Bench v2Software › Software engineering73.471.6−1.8
Excel Modeling Benchmark v1.1Professional › Finance, accounting & tax77.475.8−1.6
ARC-AGI-2Knowledge & reasoning › Abstract reasoning93.892.5−1.2
micro1 PharmaBenchProfessional › General knowledge work12.211.4−0.8
OTIS Mock AIME 2024-2025Math99.7100.0+0.3
AutomationBench-AA 1.0.6Tool & computer use › Customer service & operations74.274.4+0.3
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling1.20.7−0.2
MultiChallengeLanguage & writing › Conversation80.480.6+0.2
ARC-AGI-1 Semi-PrivateKnowledge & reasoning › Abstract reasoning98.598.50.0
Data: Epoch AI, Surge AI, Arena 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…