Atlas

Compare Models

Claude Fable 5.2 vs GPT-6 Astra

Claude Fable 5.2 leads by 3.0 capability points (≈95% interval +0.1 to +5.9): a clear lead over GPT-6 Astra · P(higher) ≈ 98%

Wins 67 of 87 shared benchmarks (17 lost, 3 tied)

Capability vs. relative cost
Claude Fable 5.2GPT-6 AstraClosed 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.

Claude Fable 5.2vsGPT-6 Astracapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins67 – 17 · 87 shared, 3 tied

Comparing matching setups only (13 benchmarks excluded) · suites counted once
← Claude Fable 5.2GPT-6 Astra →

Claude Fable 5.2vsGPT-6 Astrawins by categoryData Nov 15, 2026Data: Artificial Analysis, Epoch AI, Vals AI and 26 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-6 Astra minus Claude Fable 5.2.

HoneyBench v0.2 (hack rate)Safety11.950.8−38.8
AA-PersonalityBench Conviction indexChat & Writing75.344.8−30.8
LMArena Agent Arena Net Improvement (Steering Burden)Agents30.414.5−29.9
AA-PersonalityBench Expressiveness indexChat & Writing81.058.1−23.1
LMArena Agent Arena Praise vs ComplaintAgents73.431.7−20.9
Remote Labor Index (RLI)Professional Work41.320.8−20.5
BioSecBench-Structure V1 (58 evaluations)Science41.421.3−20.1
Humanity's Last ExamGeneral QA73.253.6−19.6
Clocktower BenchGames17591892+19.1
SWE-bench Multimodal (Vals v1, 480 Tasks)Code62.343.3−19.0
SimpleQA Verified (Epoch anti-abstention)General QA94.075.6−18.4
AA-PersonalityBench Informativeness indexChat & Writing46.729.6−17.3
Humanity's Last Exam DiamondGeneral QA80.263.8−16.4
AA-PersonalityBench Organization indexChat & Writing54.369.5+15.4
AA-3D v1.0 EloCode12711168−15.0
SpireClimbGames19.634.0+14.4
EBR-bench v4 (card ban, single agent)Games90.576.2−14.3
AA-PersonalityBench Transparency indexChat & Writing81.094.3+13.5
APEX-Agents 1.1Professional Work87.675.1−12.5
AA-Briefcase v1.1 EloAgents17851569−11.8
HardGeoBench <=10miMultimodal48.837.5−11.2
SciCode 1.0.1 (Artificial Analysis)Code67.756.5−11.2
HardGeoBench Avg DistanceMultimodal552908−11.0
GDPval-AA v2.1Professional Work18311587−10.8
OpenWorldEngineer (Mean@8)Code54.644.2−10.4
Furniture Assembly BenchmarkMultimodal90.080.0−10.0
HardGeoBench Country AccuracyMultimodal70.060.0−10.0
AA-PersonalityBench Imagination indexChat & Writing67.677.2+9.6
AA-PersonalityBench Slop indexChat & Writing6.716.2+9.6
FrontierCode 1.1 Main (Artificial Analysis)Code56.247.0−9.2
LMArena Agent Arena Steering BurdenAgents33.014.8−9.1
Finance Agent v2.1Finance & Accounting69.560.4−9.0
τ³-Banking (Mercor run, bm25_grep retrieval)Agents55.746.7−9.0
HardGeoBench <=100miMultimodal60.051.3−8.8
DAXBench v2Code88.396.8+8.5
Conceptual Reasoning IndexIndexes79.973.9−8.2
Agent Initiative BenchmarkAgents8.29.0+8.0
Vals Finance IndexIndexes63.655.6−7.9
ACCoRDGeneral QA89.481.7−7.7
AA-PersonalityBench Warmth indexChat & Writing82.975.3−7.7
FrontierSWE V2 Worst@5Code62.055.1−6.9
GDPxProfessional Work16.49.6−6.8
FrontierCode 2.0 (Artificial Analysis)Code48.842.4−6.4
Immigration BenchLegal74.067.7−6.3
LMCAGeneral QA70.664.4−6.2
FrontierSWE V2 Best@5Code79.273.0−6.2
Mystery Game PuzzlesGames90.084.0−6.0
APEX-Accounting 1.1Finance & Accounting70.464.7−5.7
AutomationBench-AA 1.0.6Agents74.268.5−5.7
Chess Puzzles (task version 2.0)Games79.074.0−5.0
GameBench 2Games99.194.3−4.8
Excel Modeling Benchmark v1.1Finance & Accounting77.472.7−4.7
BioMysteryBench (Vals Terminus 2)Science83.779.3−4.4
Epoch Capabilities Index (ECI)Indexes174169−4.1
CritPt (avg@5)Science35.731.7−4.0
CrossSearch (pass^3)Search56.060.0+4.0
LMArena Agent Arena Confirmed SuccessAgents22.014.0−4.0
Tax Agent BenchFinance & Accounting67.063.3−3.7
HieroglyphBenchMultimodal43.640.0−3.6
Ramp Accounting BenchFinance & Accounting52.248.7−3.5
AA-LCR v1.1Search83.380.0−3.3
LMArena Agent Arena Bash ReliabilityAgents14.47.8−3.3
HealthBenchHealth & Medicine60.156.9−3.2
CritPt v2 (agentic, avg@5)Science86.383.1−3.1
AutomationBench 1.0.6 (Zapier)Agents44.341.4−2.9
ChartographyMultimodal73.471.0−2.4
FrontierMath Tier 4 (v2) - PrivateMath100.097.6−2.4
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine43.941.7−2.2
HealthBench (length-adjusted, Artificial Analysis)Health & Medicine60.458.2−2.2
HLAB-Vals-1.1Legal5.07.1+2.1
GPQA DiamondGeneral QA93.995.8+1.9
DTBenchGeneral QA99.297.3−1.9
HealthBench ProfessionalHealth & Medicine71.469.7−1.8
BuseyBench SVGMultimodal7.27.3+1.5
FrontierMath Tiers 1-3 (v2) - PrivateMath95.193.7−1.4
MysteryMechanismScience51.853.2+1.4
ARC-AGI-2General QA93.895.0+1.3
ARC-AGI-3 (Semi-Private)General QA97.398.4+1.2
ProofBench v1.1Math100.099.0−1.0
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal10.09.2−0.8
ChessBenchGames69.370.0+0.7
MacroscopeBenchCode78.578.0−0.5
OTIS Mock AIME 2024-2025Math99.7100.0+0.3
DeepSWE v1.1Code74.374.1−0.2
LMArena Agent Arena Tool HallucinationAgents1.01.00.0
ARC-AGI-1 Semi-PrivateGeneral QA98.598.50.0
DualEntry Accounting AI BenchmarkFinance & Accounting81.281.20.0

Claude Fable 5.2vsGPT-6 Astranormalised score marginsData Nov 15, 2026Data: Artificial Analysis, Epoch AI, Vals AI and 26 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…