Claude Haiku 5.5 vs Gemini 4 Neon
TryTop 2 now
Gemini 4 Neon leads by 7.7 capability points (≈95% interval +5.1 to +10.4): a clear lead over Claude Haiku 5.5 · P(higher) ≈ 100%
Wins 57 of 59 shared benchmarks (2 lost)
Capability vs. relative cost
Claude Haiku 5.5Gemini 4 NeonClosed weightsOpen weights
Benchmark wins2 – 57 · 59 shared
Comparing matching setups only (3 benchmarks excluded) · suites counted once
Area← Claude Haiku 5.5Gemini 4 Neon →
Data: Artificial Analysis, Vals AI, Arena and 9 othersatlas.village.dev
Margins
Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Gemini 4 Neon minus Claude Haiku 5.5.
| Vending-Bench 2 | Business simulations | 3602 | 16843 | +78.5 |
| Mystery Game Puzzles | Games | 30.0 | 85.0 | +55.0 |
| Terminal-Bench Science 0.1 (Vals) | Science › Multidisciplinary | 10.0 | 62.9 | +52.9 |
| Terminal-Bench-Science 0.1 | Science › Multidisciplinary | 18.1 | 70.0 | +51.9 |
| SimpleQA Verified (Epoch anti-abstention) | Knowledge & reasoning › General knowledge | 23.8 | 75.2 | +51.4 |
| FrontierMath Tier 4 (v2) - Private | Math | 46.3 | 97.6 | +51.2 |
| AutomationBench 1.0.6 (Zapier) | Tool & computer use › Customer service & operations | 12.0 | 56.3 | +44.3 |
| AutomationBench-AA 1.0.6 | Tool & computer use › Customer service & operations | 36.0 | 79.3 | +43.4 |
| Humanity's Last Exam Diamond | Knowledge & reasoning › General knowledge | 29.6 | 71.3 | +41.7 |
| Furniture Assembly Benchmark | Perception & media › Spatial & visual reasoning | 47.5 | 89.2 | +41.7 |
| Bug Hunt Bench | Software › Software engineering | 20.5 | 60.0 | +39.5 |
| LMArena Agent Arena Net Improvement (Steering Burden) | Tool & computer use › General agents | -0.6 | 18.4 | +37.1 |
| SimpleBench | Knowledge & reasoning › General knowledge | 59.2 | 94.4 | +35.2 |
| Terminal-Bench 4.0 (Artificial Analysis) | Software › Software engineering | 29.3 | 63.6 | +34.3 |
| Chess Puzzles (task version 2.0) | Games | 55.0 | 83.0 | +28.0 |
| Code Migration v1.1 (Vals Index v2.2 subset) | Software › Software engineering | 50.2 | 75.8 | +25.6 |
| Terminal-Bench 4.0 | Software › Software engineering | 36.4 | 61.6 | +25.3 |
| Artificial Analysis Legal Index | Professional › Legal | 38.7 | 63.8 | +25.1 |
| LMArena Agent Arena Praise vs Complaint | Tool & computer use › General agents | -5.5 | 44.5 | +25.0 |
| Artificial Analysis Intelligence Index v5.0.1 | Composite indexes | 36.5 | 61.4 | +24.9 |
| Artificial Analysis Strategy & Ops Index | Professional › General knowledge work | 39.1 | 63.8 | +24.7 |
| Artificial Analysis Engineering Index | Science › Engineering | 42.5 | 66.8 | +24.3 |
| Vals Science Index | Science › Multidisciplinary | 38.5 | 62.6 | +24.1 |
| Artificial Analysis Healthcare & Medical Index | Health & medicine | 37.5 | 60.1 | +22.6 |
| LMArena Code Arena WebDev | Software › Frontend & design | 1596 | 1764 | +22.4 |
| Harvey LAB (Mercor run, undisclosed ~213-task cohort) | Professional › Legal | 4.9 | 26.6 | +21.7 |
| Artificial Analysis Finance & Accounting Index | Professional › Finance, accounting & tax | 43.6 | 65.1 | +21.5 |
| AA-Omniscience v1.1 Index | Knowledge & reasoning › General knowledge | 12.3 | 54.8 | +21.2 |
| APEX-Agents 1.1 | Professional › General knowledge work | 68.3 | 88.9 | +20.6 |
| Artificial Analysis Economics Index | Composite indexes | 42.7 | 61.9 | +19.2 |
| LiveBench 2.0 (2026-11-06 release) | Composite indexes | 68.9 | 87.4 | +18.5 |
| Tax Agent Bench | Professional › Finance, accounting & tax | 62.2 | 79.8 | +17.6 |
| CursorBench 4.1 | Software › Software engineering | 48.4 | 65.4 | +17.0 |
| LMArena Text Arena Elo | Language & writing › General assistance | 1421 | 1518 | +16.9 |
| τ³-Banking (Mercor run, bm25_grep retrieval) | Tool & computer use › Customer service & operations | 43.0 | 59.5 | +16.5 |
| FrontierMath Tiers 1-3 (v2) - Private | Math | 75.1 | 91.2 | +16.1 |
| MKA (Artificial Analysis) | Professional › General knowledge work | 41.0 | 56.5 | +15.5 |
| HLAB-Vals-1.1 | Professional › Legal | 4.2 | 19.6 | +15.4 |
| FrontierCode 2.0 (Artificial Analysis) | Software › Software engineering | 29.9 | 44.4 | +14.5 |
| SWE-Marathon v1.2 | Software › Software engineering | 48.1 | 62.5 | +14.4 |
| GDP.pdf (Artificial Analysis, All-pass) | Professional › General knowledge work | 20.8 | 33.4 | +12.6 |
| Legal Research Bench | Professional › Legal | 43.8 | 56.3 | +12.5 |
| Vals Index v2.2 | Composite indexes | 58.7 | 70.9 | +12.1 |
| CritPt v2 (agentic, avg@5) | Science › Physical sciences | 66.3 | 77.4 | +11.1 |
| Excel Modeling Benchmark v1.1 | Professional › Finance, accounting & tax | 65.5 | 76.5 | +11.0 |
| Vals Finance Index | Professional › Finance, accounting & tax | 57.4 | 68.3 | +10.9 |
| LMArena Agent Arena Steering Burden | Tool & computer use › General agents | 0.8 | 21.2 | +10.2 |
| MysteryMechanism | Math | 35.1 | 45.0 | +9.9 |
| AA-Briefcase v1.1 Elo | Professional › General knowledge work | 1523 | 1699 | +9.6 |
| BioMysteryBench (Vals Terminus 2) | Science › Life sciences | 70.4 | 80.0 | +9.6 |
| LMArena Agent Arena Confirmed Success | Tool & computer use › General agents | 0.2 | 19.2 | +9.5 |
| APEX-Accounting 1.1 | Professional › Finance, accounting & tax | 62.2 | 70.8 | +8.6 |
| Vals Legal Index | Professional › Legal | 70.3 | 78.4 | +8.1 |
| LMArena Agent Arena Bash Reliability | Tool & computer use › General agents | 6.6 | -6.0 | −6.3 |
| Medical Long Context Reasoning (MLCR-AA, October 2026 methodology) | Health & medicine | 35.0 | 31.7 | −3.3 |
| Vibe Code Bench v1.1 | Software › Frontend & design | 90.4 | 93.7 | +3.3 |
| Finance Agent v2.1 | Professional › Finance, accounting & tax | 65.4 | 68.6 | +3.2 |
| LMArena Agent Arena Tool Hallucination | Tool & computer use › Tool calling | -1.5 | 1.0 | +1.3 |
| OTIS Mock AIME 2024-2025 | Math | 98.9 | 100.0 | +1.1 |
Data: Artificial Analysis, Vals AI, Arena and 9 othersatlas.village.dev