Atlas

Compare Models

Claude Fable 5.2 vs Gemini 4 Carbon

Claude Fable 5.2 leads by 1.9 capability points (≈95% interval −1.2 to +5.0): statistically tied · P(higher) ≈ 89%

Wins 34 of 57 shared benchmarks (19 lost, 4 tied)

Capability vs. relative cost
Claude Fable 5.2Gemini 4 CarbonClosed 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 4 Carboncapability vs. relative costData Nov 17, 2026Village LLM Atlasatlas.village.dev

Benchmark wins34 – 19 · 57 shared, 4 tied

Comparing matching setups only (2 benchmarks excluded) · suites counted once
← Claude Fable 5.2Gemini 4 Carbon →

Claude Fable 5.2vsGemini 4 Carbonwins by categoryData Nov 17, 2026Data: Epoch AI, Artificial Analysis, Arena and 17 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Gemini 4 Carbon minus Claude Fable 5.2.

Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine43.919.4−24.4
LMArena Agent Arena Net Improvement (Steering Burden)Agents28.616.4−23.2
OpenWorldEngineer (Mean@8)Code54.632.8−21.8
LMArena Agent Arena Praise vs ComplaintAgents76.040.3−17.9
Mystery Game PuzzlesGames90.073.0−17.0
SimpleQA Verified (Epoch anti-abstention)General QA94.077.7−16.3
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal10.023.0+13.0
Furniture Assembly BenchmarkMultimodal90.077.5−12.5
HLAB-Vals-1.1Legal5.017.5+12.5
FrontierSWE V2 Best@5Code79.266.7−12.5
Humanity's Last Exam DiamondGeneral QA80.269.1−11.1
Vals RSI Index v1.1 (four tasks)Agents37.927.7−10.1
LMArena Agent Arena Bash ReliabilityAgents13.6-5.9−9.8
Tax Agent BenchFinance & Accounting67.076.4+9.4
BrokenArXiv 10/2026 (Native Harnesses)Math73.582.7+9.3
AutomationBench 1.0.6 (Zapier)Agents44.353.0+8.7
FrontierCode 2.0 - Pass RateCode54.746.9−7.8
LMArena Code Arena WebDevCode18261768−7.6
FrontierCode 2.0 - ScoreCode50.242.6−7.6
BuseyBench SVGMultimodal7.27.9+6.9
FrontierSWE V2 Worst@5Code62.056.2−5.9
GBENCH One-shot Coding GScoreGames74.969.0−5.9
GDPxProfessional Work16.410.9−5.6
DAXBench v2Code88.393.6+5.3
FrontierMath Tier 4 (v2) - PrivateMath100.095.1−4.9
Vibe Code Bench v1.1Code93.889.1−4.7
FrontierMath Tiers 1-3 (v2) - PrivateMath95.190.5−4.6
AA-LCR v1.1Search83.379.3−4.0
GDPval v2 (OpenAI internal)Professional Work44.848.8+4.0
LMArena Agent Arena Steering BurdenAgents25.518.1−3.7
AutomationBench-AA 1.0.6Agents74.277.8+3.7
AA-Briefcase v1.1 EloAgents16771611−3.6
Vals Finance IndexIndexes63.667.0+3.5
Epoch Capabilities Index (ECI)Indexes174170−3.0
EBR-bench v4 (card ban, single agent)Games90.587.6−2.9
APEX-Agents 1.1Professional Work87.690.4+2.8
GPQA DiamondGeneral QA93.996.5+2.7
Finance Agent v2.1Finance & Accounting69.567.2−2.3
ARC-AGI-3 (Semi-Private)General QA96.899.1+2.3
Blueprint-Bench 2Multimodal0.60.6+2.0
BrokenArXiv 09/2026 (Native Harnesses)Math80.582.4+1.9
CritPt v2 (agentic, avg@5)Science83.782.3−1.4
Chess Puzzles (task version 2.0)Games79.078.0−1.0
Excel Modeling Benchmark v1.1Finance & Accounting77.476.5−0.9
ChartographyMultimodal73.472.5−0.9
LMArena Agent Arena Confirmed SuccessAgents19.417.9−0.8
GradeBenchProfessional Work57.356.6−0.7
Bug Hunt BenchCode55.254.6−0.6
GDPval-AA v2.1Professional Work17181733+0.6
ARC-AGI-2General QA93.894.2+0.4
APEX-Accounting 1.1Finance & Accounting70.470.7+0.3
OTIS Mock AIME 2024-2025Math99.7100.0+0.3
LMArena Agent Arena Tool HallucinationAgents1.00.6−0.2
ARC-AGI-1 Semi-PrivateGeneral QA98.598.50.0
ArXivMath 09/2026 (Native Harnesses)Math96.296.20.0
ArXivMath 10/2026 (Native Harnesses)Math93.693.60.0
τ³-Banking (Mercor run, bm25_grep retrieval)Agents55.755.70.0

Claude Fable 5.2vsGemini 4 Carbonnormalised score marginsData Nov 17, 2026Data: Epoch AI, Artificial Analysis, Arena and 17 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…