Atlas

Compare Models

Claude Fable 5.2 vs Gemini 3.8 Flash

Claude Fable 5.2 leads by 9.6 capability points (≈95% interval +6.9 to +12.3): a clear lead over Gemini 3.8 Flash · P(higher) ≈ 100%

Wins 71 of 82 shared benchmarks (10 lost, 1 tied)

Capability vs. relative cost
Claude Fable 5.2Gemini 3.8 FlashClosed 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 5.2vsGemini 3.8 Flashcapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins71 – 10 · 82 shared, 1 tied

Comparing matching setups only (5 benchmarks excluded) · suites counted once
← Claude Fable 5.2Gemini 3.8 Flash →

Claude Fable 5.2vsGemini 3.8 Flashwins by categoryData Nov 15, 2026Data: Artificial Analysis, Vals AI, Epoch AI and 24 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Gemini 3.8 Flash minus Claude Fable 5.2.

FrontierMath Tier 4 (v2) - PrivateMath100.022.0−78.0
ARC-AGI-3 (Semi-Private)General QA96.824.2−72.6
HoneyBench v0.2 (hack rate)Safety11.980.5−68.6
AA-PersonalityBench Slop indexChat & Writing6.771.4+65.4
Furniture Assembly BenchmarkMultimodal90.031.7−58.3
AA-PersonalityBench Transparency indexChat & Writing81.023.8−57.7
BrokenArXiv 09/2026 (Native Harnesses)Math80.526.7−53.8
LMArena Agent Arena Net Improvement (Steering Burden)Agents30.42.3−53.0
ProofBench v1.1Math100.048.0−52.0
Humanity's Last Exam DiamondGeneral QA80.229.1−51.1
FrontierSWE V2Code69.919.6−50.3
ArXivMath 09/2026 (Native Harnesses)Math96.250.0−46.2
AA-3D v1.0 EloCode1271974−43.2
Mystery Game PuzzlesGames90.047.0−43.0
Bug Hunt BenchCode55.217.1−38.1
Remote Labor Index (RLI)Professional Work41.35.8−35.5
GameBench 2Games99.164.0−35.1
AA-PersonalityBench Informativeness indexChat & Writing46.781.0+34.6
LMArena Agent Arena Praise vs ComplaintAgents73.44.9−34.3
LMArena Code Arena WebDevCode18361583−33.1
ChartographyMultimodal73.440.9−32.5
FrontierCode 2.0 (Artificial Analysis)Code49.118.7−30.4
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine43.915.6−28.3
CrossSearch (pass^3)Search56.029.0−27.0
Humanity's Last ExamGeneral QA73.246.2−27.0
AA-PersonalityBench Imagination indexChat & Writing67.641.0−26.9
FrontierCode 1.1 Main (Artificial Analysis)Code64.037.1−26.9
FrontierMath Tiers 1-3 (v2) - PrivateMath95.168.4−26.7
AA-Briefcase v1.1 EloAgents16771204−25.9
SimpleQA Verified (Epoch anti-abstention)General QA94.069.7−24.3
AA-PersonalityBench Warmth indexChat & Writing82.960.0−23.1
BioMysteryBench (Vals Terminus 2)Science83.762.2−21.5
SWE-bench Multimodal (Vals v1, 480 Tasks)Code62.340.8−21.5
Chess Puzzles (task version 2.0)Games79.058.0−21.0
Conceptual Reasoning IndexIndexes79.965.7−19.5
AA-PersonalityBench Organization indexChat & Writing54.373.3+19.2
CritPt v2 (agentic, avg@5)Science83.765.1−18.6
Vals RSI Index v1.1 (four tasks)Agents37.919.9−18.0
LMCAGeneral QA70.652.9−17.7
CritPt (avg@5)Science35.718.3−17.4
LMArena Agent Arena Steering BurdenAgents33.0-1.3−17.1
MysteryMechanismScience51.836.5−15.3
Vibe Code Bench v1.1Code93.878.7−15.1
APEX-Agents 1.1Professional Work87.672.9−14.7
AutomationBench 1.0.6 (Zapier)Agents44.329.7−14.6
AutomationBench-AA 1.0.6Agents74.260.9−13.3
Epoch Capabilities Index (ECI)Indexes174159−12.6
HealthBench ProfessionalHealth & Medicine71.459.0−12.4
GDPval-AA v2.1Professional Work17181442−12.2
HealthBench (length-adjusted, Artificial Analysis)Health & Medicine60.448.4−12.0
APEX-Accounting 1.1Finance & Accounting70.458.8−11.6
ACCoRDGeneral QA89.478.2−11.2
SciCode 1.0.1 (Artificial Analysis)Code67.756.6−11.1
HieroglyphBenchMultimodal43.654.6+11.0
SWE-bench MultilingualCode96.685.8−10.8
Immigration BenchLegal74.063.4−10.6
LMArena Agent Arena Confirmed SuccessAgents22.01.3−10.3
AA-PersonalityBench Conviction indexChat & Writing75.384.8+9.6
MineBench — Bradley-Terry RatingGames19321806−7.6
HLAB-Vals-1.1Legal5.012.1+7.1
τ³-Banking (Mercor run, bm25_grep retrieval)Agents55.748.8−6.9
Ramp Accounting BenchFinance & Accounting52.245.7−6.4
BuseyBench SVGMultimodal7.26.6−5.4
ARC-AGI-2General QA93.889.2−4.6
Excel Modeling Benchmark v1.1Finance & Accounting77.472.9−4.5
ChessBenchGames69.373.3+4.0
Finance Agent v2.1Finance & Accounting69.565.8−3.7
DTBenchGeneral QA99.295.7−3.5
DeepSWE v1.1Code74.371.0−3.3
Vals Finance IndexIndexes63.660.5−3.1
HealthBenchHealth & Medicine60.157.1−3.0
LMArena Agent Arena Bash ReliabilityAgents14.49.9−2.3
DualEntry Accounting AI BenchmarkFinance & Accounting81.283.2+2.0
MacroscopeBenchCode72.770.7−2.0
AA-PersonalityBench Expressiveness indexChat & Writing81.079.1−1.9
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal10.011.4+1.4
DAXBench v2Code88.387.2−1.1
OTIS Mock AIME 2024-2025Math99.798.9−0.8
AA-LCR v1.1Search83.384.0+0.7
LMArena Agent Arena Tool HallucinationAgents1.00.4−0.3
Tax Agent BenchFinance & Accounting67.066.8−0.2
ARC-AGI-1 Semi-PrivateGeneral QA98.598.50.0

Claude Fable 5.2vsGemini 3.8 Flashnormalised score marginsData Nov 15, 2026Data: Artificial Analysis, Vals AI, Epoch AI and 24 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…