Gemini 4 Carbon vs Gemini 4 Neon
TryTop 2 now
Gemini 4 Neon leads by 1.3 capability points (≈95% interval −1.7 to +4.2): statistically tied · P(higher) ≈ 80%
Wins 56 of 78 shared benchmarks (20 lost, 2 tied)
Capability vs. relative cost
Gemini 4 CarbonGemini 4 NeonClosed weightsOpen weights
Benchmark wins20 – 56 · 78 shared, 2 tied
Comparing matching setups only · suites counted once
Area← Gemini 4 CarbonGemini 4 Neon →
Data: Artificial Analysis, Surge AI, Vals AI and 14 othersatlas.village.dev
Margins
Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Gemini 4 Neon minus Gemini 4 Carbon.
| DAYJOB: Finance (October 2026 rescoring) | Professional › Finance, accounting & tax | 23.8 | 50.5 | +26.7 |
| DAYJOB: Accounting | Professional › Finance, accounting & tax | 25.2 | 49.6 | +24.4 |
| DAYJOB: Law (November 2026 release) | Professional › Legal | 13.8 | 26.2 | +12.4 |
| Medical Long Context Reasoning (MLCR-AA, October 2026 methodology) | Health & medicine | 19.4 | 31.7 | +12.2 |
| DAYJOB: Healthcare (October 2026 rescoring) | Health & medicine | 35.2 | 47.2 | +12.0 |
| Mystery Game Puzzles | Games | 73.0 | 85.0 | +12.0 |
| Furniture Assembly Benchmark | Perception & media › Spatial & visual reasoning | 77.5 | 89.2 | +11.7 |
| DAYJOB: Tax | Professional › Finance, accounting & tax | 4.8 | 12.8 | +8.0 |
| sudo L7 (October 2026 rescoring) | Software › Software engineering | 30.6 | 38.3 | +7.7 |
| Terminal-Bench-Science 0.1 | Science › Multidisciplinary | 62.4 | 54.8 | −7.6 |
| Vals RSI Index v1.1 (four tasks) | Software › Data & ML | 27.7 | 34.4 | +6.7 |
| ExploitGym - release-reported percentage, cohort unspecified | Software › Cybersecurity | 54.8 | 48.3 | −6.5 |
| Terminal-Bench 4.0 (Artificial Analysis) | Software › Software engineering | 60.6 | 54.5 | −6.1 |
| Chess Puzzles (task version 2.0) | Games | 78.0 | 83.0 | +5.0 |
| Humanity's Sixth Sense | Perception & media › Spatial & visual reasoning | 71.1 | 75.9 | +4.8 |
| Vibe Code Bench v1.1 | Software › Frontend & design | 89.1 | 93.7 | +4.6 |
| LMArena Agent Arena Net Improvement (Steering Burden) | Tool & computer use › General agents | 16.1 | 18.4 | +4.5 |
| Code Migration v1.1 (Vals Index v2.2 subset) | Software › Software engineering | 80.1 | 75.8 | −4.3 |
| CritPt v2 (agentic, avg@5) | Science › Physical sciences | 82.3 | 78.0 | −4.3 |
| GDP.xlsx | Professional › General knowledge work | 38.9 | 43.1 | +4.2 |
| SWE-Marathon v1.2 | Software › Software engineering | 66.7 | 62.5 | −4.2 |
| LiveBench 2.0 (2026-11-06 release) | Composite indexes | 83.3 | 87.4 | +4.1 |
| RiemannBench (November 2026 rescoring) | Math | 80.0 | 84.0 | +4.0 |
| CursorBench 4.1 | Software › Software engineering | 61.5 | 65.4 | +3.9 |
| τ³-Banking (Mercor run, bm25_grep retrieval) | Tool & computer use › Customer service & operations | 55.7 | 59.5 | +3.8 |
| GDP.pdf (Artificial Analysis, All-pass) | Professional › General knowledge work | 29.6 | 33.4 | +3.8 |
| Harvey LAB (Mercor run, undisclosed ~213-task cohort) | Professional › Legal | 23.0 | 26.6 | +3.6 |
| CWE-bench v2 Pass@1 | Software › Cybersecurity | 67.3 | 70.7 | +3.4 |
| Tax Agent Bench | Professional › Finance, accounting & tax | 76.4 | 79.8 | +3.4 |
| AutomationBench 1.0.6 (Zapier) | Tool & computer use › Customer service & operations | 53.0 | 56.3 | +3.4 |
| Artificial Analysis Healthcare & Medical Index | Health & medicine | 56.8 | 60.1 | +3.3 |
| SWE-sweep | Software › Software engineering | 9.6 | 12.8 | +3.2 |
| Artificial Analysis Coding Agent Index v2.1 | Software › Software engineering | 55.5 | 58.6 | +3.2 |
| EnterpriseBench: CoreCraft | Tool & computer use › Customer service & operations | 80.5 | 77.4 | −3.1 |
| GDP.pdf (November 2026 rescoring) | Professional › General knowledge work | 56.4 | 53.4 | −3.0 |
| Artificial Analysis Strategy & Ops Index | Professional › General knowledge work | 60.9 | 63.8 | +2.8 |
| FrontierCode 2.0 (Artificial Analysis) | Software › Software engineering | 40.6 | 43.3 | +2.7 |
| LMArena Agent Arena Praise vs Complaint | Tool & computer use › General agents | 39.4 | 44.5 | +2.5 |
| HANDBOOK.md Agents v1.1 | Tool & computer use › Customer service & operations | 49.8 | 52.3 | +2.5 |
| SimpleQA Verified (Epoch anti-abstention) | Knowledge & reasoning › General knowledge | 77.7 | 75.2 | −2.5 |
| FrontierMath Tier 4 (v2) - Private | Math | 95.1 | 97.6 | +2.4 |
| Artificial Analysis Finance & Accounting Index | Professional › Finance, accounting & tax | 62.6 | 65.1 | +2.4 |
| Vals Legal Index | Professional › Legal | 75.9 | 78.4 | +2.4 |
| Artificial Analysis Economics Index | Composite indexes | 59.6 | 61.9 | +2.3 |
| SEC-Bench Pro - release-reported cohort | Software › Cybersecurity | 88.2 | 90.5 | +2.3 |
| Humanity's Last Exam Diamond | Knowledge & reasoning › General knowledge | 69.1 | 71.3 | +2.2 |
| HLAB-Vals-1.1 | Professional › Legal | 17.5 | 19.6 | +2.1 |
| ComplexConstraints (November 2026 rescoring) | Language & writing › Instruction following | 70.9 | 72.9 | +2.0 |
| SimpleBench | Knowledge & reasoning › General knowledge | 92.6 | 94.4 | +1.8 |
| Artificial Analysis Legal Index | Professional › Legal | 62.0 | 63.8 | +1.8 |
| LMArena Agent Arena Steering Burden | Tool & computer use › General agents | 17.9 | 21.2 | +1.6 |
| Agents' Last Exam v1.1 Score | Professional › General knowledge work | 77.4 | 75.8 | −1.6 |
| APEX-Agents 1.1 | Professional › General knowledge work | 90.4 | 88.8 | −1.6 |
| Artificial Analysis Engineering Index | Science › Engineering | 65.3 | 66.8 | +1.5 |
| AutomationBench-AA 1.0.6 | Tool & computer use › Customer service & operations | 77.8 | 79.3 | +1.5 |
| Terminal-Bench 4.0 | Software › Software engineering | 63.1 | 61.6 | −1.5 |
| Finance Agent v2.1 | Professional › Finance, accounting & tax | 67.2 | 68.6 | +1.4 |
| Artificial Analysis Intelligence Index v5.0.1 | Composite indexes | 60.0 | 61.4 | +1.4 |
| LMArena Text Arena Elo | Language & writing › General assistance | 1526 | 1518 | −1.3 |
| Vals Finance Index | Professional › Finance, accounting & tax | 67.0 | 68.3 | +1.3 |
| AA-Briefcase v1.1 Elo | Professional › General knowledge work | 1611 | 1632 | +1.2 |
| APEX-1 (Mercor, mean score) | Professional › General knowledge work | 80.0 | 78.9 | −1.1 |
| LMArena Agent Arena Confirmed Success | Tool & computer use › General agents | 17.2 | 19.2 | +1.0 |
| Chartography | Perception & media › Documents & charts | 72.5 | 73.5 | +1.0 |
| GraphWalks BFS: 256K to 1M Aggregate | Knowledge & reasoning › Reading & long context | 96.2 | 95.2 | −1.0 |
| FrontierMath Tiers 1-3 (v2) - Private | Math | 90.5 | 91.2 | +0.7 |
| AA-Omniscience v1.1 Index | Knowledge & reasoning › General knowledge | 53.5 | 54.8 | +0.6 |
| LMArena Code Arena WebDev | Software › Frontend & design | 1769 | 1764 | −0.6 |
| Legal Research Bench | Professional › Legal | 55.8 | 56.3 | +0.5 |
| Bug Hunt Bench | Software › Software engineering | 54.6 | 54.3 | −0.3 |
| GPQA Diamond | Science › Multidisciplinary | 96.5 | 96.3 | −0.3 |
| MKA (Artificial Analysis) | Professional › General knowledge work | 54.3 | 54.6 | +0.3 |
| LMArena Agent Arena Bash Reliability | Tool & computer use › General agents | -5.7 | -6.0 | −0.1 |
| APEX-Accounting 1.1 | Professional › Finance, accounting & tax | 70.7 | 70.8 | +0.1 |
| LMArena Agent Arena Tool Hallucination | Tool & computer use › Tool calling | 0.8 | 1.0 | +0.1 |
| Excel Modeling Benchmark v1.1 | Professional › Finance, accounting & tax | 76.5 | 76.5 | 0.0 |
| Gray Swan IPI Attack Success (k=15) | Safety & alignment | 0.0 | 0.0 | 0.0 |
| OTIS Mock AIME 2024-2025 | Math | 100.0 | 100.0 | 0.0 |
Data: Artificial Analysis, Surge AI, Vals AI and 14 othersatlas.village.dev