Atlas

Compare Models

Claude Fable 5.2 vs GPT-6.2 Astra

Claude Fable 5.2 leads by 1.3 capability points (≈95% interval −1.7 to +4.2): statistically tied · P(higher) ≈ 80%

Wins 45 of 73 shared benchmarks (25 lost, 3 tied)

Capability vs. relative cost
Claude Fable 5.2GPT-6.2 AstraClosed weightsOpen weights

Benchmark wins45 – 25 · 73 shared, 3 tied

Comparing matching setups only (46 benchmarks excluded) · suites counted once
← Claude Fable 5.2GPT-6.2 Astra →
Data: Artificial Analysis, Surge AI, Epoch AI and 17 othersatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-6.2 Astra minus Claude Fable 5.2.

KradleBenchGames37.878.3+40.5
AA-PersonalityBench Transparency indexLanguage & writing › Style & personality81.040.9−40.4
AA-PersonalityBench Organization indexLanguage & writing › Style & personality54.321.8−32.9
SpireClimbGames15.341.3+26.0
EnterpriseBench: CoreCraftTool & computer use › Customer service & operations68.243.6−24.6
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents27.415.2−23.8
SWE Atlas v1.1 - OverallSoftware › Software engineering80.560.5−20.1
TaxCalcBench TY25Professional › Finance, accounting & tax62.044.0−18.0
SimpleBenchKnowledge & reasoning › General knowledge94.977.3−17.6
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents68.737.0−15.8
AA-PersonalityBench Informativeness indexLanguage & writing › Style & personality46.731.3−15.5
SimpleQA Verified (Epoch anti-abstention)Knowledge & reasoning › General knowledge94.078.5−15.5
AA-PersonalityBench Warmth indexLanguage & writing › Style & personality82.968.1−15.0
DAYJOB: Law (November 2026 release)Professional › Legal26.541.4+14.9
AA-PersonalityBench Expressiveness indexLanguage & writing › Style & personality81.066.5−14.6
sudo L7 (October 2026 rescoring)Software › Software engineering52.238.3−13.9
HLAB-Vals-1.1Professional › Legal5.016.3+11.3
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine45.034.4−10.6
Chartography (November 2026 rescoring)Perception & media › Documents & charts80.890.7+9.9
Tax Agent BenchProfessional › Finance, accounting & tax67.057.9−9.1
Finance Agent v2.1Professional › Finance, accounting & tax69.560.7−8.7
AutomationBench 1.0.6 (Zapier)Tool & computer use › Customer service & operations44.352.4+8.1
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge80.272.2−8.0
AA-Briefcase v1.1 EloProfessional › General knowledge work17851644−7.7
MultiChallengeLanguage & writing › Conversation80.473.2−7.2
RiemannBench (November 2026 rescoring)Math83.276.0−7.2
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations52.345.8−6.5
DAYJOB: TaxProfessional › Finance, accounting & tax15.69.2−6.4
AA-3D v1.0 EloSoftware › Frontend & design12711318+6.3
Vals Finance IndexProfessional › Finance, accounting & tax63.657.3−6.3
Professional Reasoning Benchmark - LegalProfessional › Legal49.043.1−5.9
GDPval v2 (OpenAI internal)Professional › General knowledge work44.850.4+5.6
Agents' Last Exam v1.1 ScoreProfessional › General knowledge work72.878.3+5.5
LMArena Agent Arena Steering BurdenTool & computer use › General agents24.514.0−5.3
ComplexConstraints (November 2026 rescoring)Language & writing › Instruction following64.569.6+5.1
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents22.512.4−5.0
Dual-Arm Sim (task completion)Tool & computer use › Robotics7.52.5−5.0
Mystery Game PuzzlesGames90.095.0+5.0
Blueprint-Bench 2Perception & media › Spatial & visual reasoning0.60.6+4.8
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine43.939.4−4.4
AA-PersonalityBench Imagination indexLanguage & writing › Style & personality67.663.3−4.4
AA-PersonalityBench Conviction indexLanguage & writing › Style & personality75.371.3−4.0
Vibe Code Bench v1.1Software › Frontend & design93.889.9−3.9
Humanity's Sixth SensePerception & media › Spatial & visual reasoning59.563.2+3.7
DAYJOB: AccountingProfessional › Finance, accounting & tax45.242.0−3.2
GDP.xlsxProfessional › General knowledge work32.929.7−3.2
AutomationBench-AA 1.0.6Tool & computer use › Customer service & operations74.277.2+3.1
DualEntry Accounting AI BenchmarkProfessional › Finance, accounting & tax81.278.2−3.0
GPQA DiamondScience › Multidisciplinary93.996.8+3.0
BioMysteryBench (Vals Terminus 2)Science › Life sciences83.780.7−3.0
ARC-AGI-3 (Semi-Private)Games97.3100.0+2.7
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents13.17.7−2.7
GDP.pdf (November 2026 rescoring)Professional › General knowledge work51.849.2−2.6
CritPt v2 (agentic, avg@5)Science › Physical sciences86.383.7−2.6
FrontierMath Tiers 1-3 (v2) - PrivateMath95.192.6−2.5
LLM Debate BenchmarkLanguage & writing › Language & wordplay16881673−2.1
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax43.745.8+2.1
ARC-AGI-2Knowledge & reasoning › Abstract reasoning93.895.8+2.1
Ramp SWE-Bench v2 (Bash harness)Software › Software engineering73.075.0+2.0
EBR-bench v4 (card ban, single agent)Games90.592.4+1.9
Gray Swan IPI Attack Success (k=15)Safety & alignment4.12.4+1.7
FrontierCode 2.0 (Artificial Analysis)Software › Software engineering47.649.0+1.4
MysteryMechanismMath51.853.2+1.4
Excel Modeling Benchmark v1.1Professional › Finance, accounting & tax77.476.0−1.3
Chess Puzzles (task version 2.0)Games79.078.0−1.0
Furniture Assembly BenchmarkPerception & media › Spatial & visual reasoning90.089.2−0.8
AA-PersonalityBench Slop indexLanguage & writing › Style & personality6.77.4+0.7
Epoch Capabilities Index (ECI)Composite indexes173173−0.5
OTIS Mock AIME 2024-2025Math99.7100.0+0.3
LiveBench 2.0 (2026-11-06 release)Composite indexes84.984.90.0
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling1.21.20.0
ARC-AGI-1 Semi-PrivateKnowledge & reasoning › Abstract reasoning98.598.50.0
FrontierMath Tier 4 (v2) - PrivateMath100.0100.00.0
Data: Artificial Analysis, Surge AI, Epoch AI and 17 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…