Claude Fable 5 vs Gemini 4 Carbon
TryTop 2 now
Gemini 4 Carbon leads by 4.5 capability points (≈95% interval +1.8 to +7.1): a clear lead over Claude Fable 5 · P(higher) ≈ 100%
Wins 36 of 43 shared benchmarks (5 lost, 2 tied)
Capability vs. relative cost
Claude Fable 5Gemini 4 CarbonClosed weightsOpen weights
Benchmark wins5 – 36 · 43 shared, 2 tied
Comparing matching setups only (8 benchmarks excluded) · suites counted once
Area← Claude Fable 5Gemini 4 Carbon →
Data: Epoch AI, Arena, Surge AI and 14 othersatlas.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.
| HieroglyphBench | Perception & media › Documents & charts | 22.9 | 76.0 | +53.1 |
| Furniture Assembly Benchmark | Perception & media › Spatial & visual reasoning | 35.8 | 77.5 | +41.7 |
| Chartography | Perception & media › Documents & charts | 34.8 | 72.5 | +37.7 |
| Humanity's Sixth Sense | Perception & media › Spatial & visual reasoning | 34.5 | 71.1 | +36.6 |
| AutomationBench 1.0.6 (Zapier) | Tool & computer use › Customer service & operations | 17.1 | 53.0 | +35.9 |
| Chess Puzzles (task version 2.0) | Games | 47.0 | 78.0 | +31.0 |
| FrontierCUA v1 pass@1 | Tool & computer use › Computer & browser use | 15.0 | 42.5 | +27.5 |
| Bug Hunt Bench | Software › Software engineering | 27.6 | 54.6 | +27.0 |
| DAYJOB: Healthcare (October 2026 rescoring) | Health & medicine | 11.2 | 35.2 | +24.0 |
| Humanity's Last Exam Diamond | Knowledge & reasoning › General knowledge | 45.6 | 69.1 | +23.5 |
| Mystery Game Puzzles | Games | 52.0 | 73.0 | +21.0 |
| Terminal-Bench 4.0 (Artificial Analysis) | Software › Software engineering | 42.4 | 62.6 | +20.2 |
| RiemannBench | Math | 60.0 | 80.0 | +20.0 |
| Blueprint-Bench 2 | Perception & media › Spatial & visual reasoning | 0.4 | 0.6 | +20.0 |
| LMArena Agent Arena Net Improvement (Steering Burden) | Tool & computer use › General agents | 5.9 | 16.1 | +20.0 |
| LMArena Code Arena WebDev | Software › Frontend & design | 1626 | 1769 | +19.0 |
| MMMU Pro | Knowledge & reasoning › General knowledge | 72.7 | 89.2 | +16.5 |
| APEX-Agents 1.1 | Professional › General knowledge work | 74.5 | 90.4 | +15.9 |
| LMArena Agent Arena Praise vs Complaint | Tool & computer use › General agents | 11.5 | 39.4 | +14.0 |
| GPQA Diamond | Science › Multidisciplinary | 83.3 | 96.5 | +13.2 |
| EnterpriseBench: CoreCraft | Tool & computer use › Customer service & operations | 70.3 | 80.5 | +10.2 |
| APEX-Accounting 1.1 | Professional › Finance, accounting & tax | 61.0 | 70.7 | +9.7 |
| LMArena Agent Arena Bash Reliability | Tool & computer use › General agents | 9.0 | -5.7 | −7.4 |
| Tax Agent Bench | Professional › Finance, accounting & tax | 69.8 | 76.4 | +6.6 |
| Gray Swan IPI Attack Success (k=15) | Safety & alignment | 6.5 | 0.0 | +6.5 |
| LMArena Agent Arena Confirmed Success | Tool & computer use › General agents | 4.4 | 17.2 | +6.4 |
| LMArena Agent Arena Steering Burden | Tool & computer use › General agents | 5.5 | 17.9 | +6.2 |
| HiL-Bench (Human-in-Loop Benchmark) | Tool & computer use › General agents | 56.3 | 50.3 | −6.0 |
| DAYJOB: Finance (October 2026 rescoring) | Professional › Finance, accounting & tax | 18.3 | 23.8 | +5.5 |
| GBENCH One-shot Coding GScore | Games | 63.6 | 69.0 | +5.4 |
| ARC-AGI-2 | Knowledge & reasoning › Abstract reasoning | 89.2 | 94.2 | +5.0 |
| FrontierMath Tier 4 (v2) - Private | Math | 90.2 | 95.1 | +4.9 |
| LLM Debate Benchmark | Language & writing › Language & wordplay | 1705 | 1676 | −4.1 |
| AA-Briefcase v1.1 Elo | Professional › General knowledge work | 1536 | 1611 | +4.1 |
| LMArena Text Arena Elo | Language & writing › General assistance | 1504 | 1526 | +3.8 |
| FrontierMath Tiers 1-3 (v2) - Private | Math | 87.0 | 90.5 | +3.5 |
| Vals RSI Index v1.1 (four tasks) | Software › Data & ML | 24.3 | 27.7 | +3.4 |
| FrontierMath Erdős | Math | 0.0 | 2.9 | +2.9 |
| Vibe Code Bench v1.1 | Software › Frontend & design | 90.4 | 89.1 | −1.3 |
| Medical Long Context Reasoning (MLCR-AA, October 2026 methodology) | Health & medicine | 20.6 | 19.4 | −1.1 |
| LMArena Agent Arena Tool Hallucination | Tool & computer use › Tool calling | 0.5 | 0.8 | +0.2 |
| ARC-AGI-1 Semi-Private | Knowledge & reasoning › Abstract reasoning | 98.5 | 98.5 | 0.0 |
| OTIS Mock AIME 2024-2025 | Math | 100.0 | 100.0 | 0.0 |
Data: Epoch AI, Arena, Surge AI and 14 othersatlas.village.dev