Gemini 4 Neon vs GPT-6.2 Astra
Gemini 4 Neon leads by 0.4 capability points (≈95% interval −2.6 to +3.4): statistically tied · P(higher) ≈ 61%
Wins 40 of 66 shared benchmarks (23 lost, 3 tied)
Capability vs. relative cost
Gemini 4 NeonGPT-6.2 AstraClosed weightsOpen weights
Benchmark wins40 – 23 · 66 shared, 3 tied
Comparing matching setups only (34 benchmarks excluded) · suites counted once
Area← Gemini 4 NeonGPT-6.2 Astra →
Data: Artificial Analysis, Vals AI, Surge AI and 8 othersatlas.village.dev
Margins
Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-6.2 Astra minus Gemini 4 Neon.
| EnterpriseBench: CoreCraft | Tool & computer use › Customer service & operations | 77.4 | 43.6 | −33.8 |
| Tax Agent Bench | Professional › Finance, accounting & tax | 79.8 | 57.9 | −21.9 |
| SimpleBench | Knowledge & reasoning › General knowledge | 94.4 | 77.3 | −17.1 |
| DAYJOB: Law (November 2026 release) | Professional › Legal | 26.2 | 41.4 | +15.2 |
| Legal Research Bench | Professional › Legal | 56.3 | 42.8 | −13.5 |
| GDP.xlsx | Professional › General knowledge work | 43.1 | 29.7 | −13.4 |
| DAYJOB: Healthcare (October 2026 rescoring) | Health & medicine | 47.2 | 34.4 | −12.8 |
| Humanity's Sixth Sense | Perception & media › Spatial & visual reasoning | 75.9 | 63.2 | −12.6 |
| Vals Finance Index | Professional › Finance, accounting & tax | 68.3 | 57.3 | −11.0 |
| Mystery Game Puzzles | Games | 85.0 | 95.0 | +10.0 |
| Vals Legal Index | Professional › Legal | 78.4 | 69.7 | −8.7 |
| Terminal-Bench Science 0.1 (Vals) | Science › Multidisciplinary | 62.9 | 71.4 | +8.6 |
| Terminal-Bench-Science 0.1 | Science › Multidisciplinary | 70.0 | 78.6 | +8.6 |
| MysteryMechanism | Math | 45.0 | 53.2 | +8.1 |
| RiemannBench (November 2026 rescoring) | Math | 84.0 | 76.0 | −8.0 |
| Code Migration v1.1 (Vals Index v2.2 subset) | Software › Software engineering | 75.8 | 83.6 | +7.8 |
| Finance Agent v2.1 | Professional › Finance, accounting & tax | 68.6 | 60.7 | −7.8 |
| Medical Long Context Reasoning (MLCR-AA, October 2026 methodology) | Health & medicine | 31.7 | 39.4 | +7.8 |
| DAYJOB: Accounting | Professional › Finance, accounting & tax | 49.6 | 42.0 | −7.6 |
| LMArena Agent Arena Bash Reliability | Tool & computer use › General agents | -6.0 | 7.7 | +6.8 |
| HANDBOOK.md Agents v1.1 | Tool & computer use › Customer service & operations | 52.3 | 45.8 | −6.5 |
| LMArena Agent Arena Net Improvement (Steering Burden) | Tool & computer use › General agents | 18.4 | 15.2 | −6.2 |
| Vals Science Index | Science › Multidisciplinary | 62.6 | 68.4 | +5.8 |
| AA-Briefcase v1.1 Elo | Professional › General knowledge work | 1699 | 1594 | −5.7 |
| CritPt v2 (agentic, avg@5) | Science › Physical sciences | 77.4 | 83.1 | +5.7 |
| Artificial Analysis Legal Index | Professional › Legal | 63.8 | 58.7 | −5.0 |
| Chess Puzzles (task version 2.0) | Games | 83.0 | 78.0 | −5.0 |
| DAYJOB: Finance (October 2026 rescoring) | Professional › Finance, accounting & tax | 50.5 | 45.8 | −4.7 |
| FrontierCode 2.0 (Artificial Analysis) | Software › Software engineering | 44.4 | 49.0 | +4.5 |
| GDP.pdf (November 2026 rescoring) | Professional › General knowledge work | 53.4 | 49.2 | −4.2 |
| Artificial Analysis Finance & Accounting Index | Professional › Finance, accounting & tax | 65.1 | 60.9 | −4.1 |
| Chartography (November 2026 rescoring) | Perception & media › Documents & charts | 86.6 | 90.7 | +4.1 |
| AutomationBench 1.0.6 (Zapier) | Tool & computer use › Customer service & operations | 56.3 | 52.4 | −4.0 |
| MKA (Artificial Analysis) | Professional › General knowledge work | 56.5 | 52.7 | −3.8 |
| Vibe Code Bench v1.1 | Software › Frontend & design | 93.7 | 89.9 | −3.8 |
| SEC-Bench Pro - release-reported cohort | Software › Cybersecurity | 90.5 | 86.8 | −3.7 |
| LMArena Agent Arena Praise vs Complaint | Tool & computer use › General agents | 44.5 | 37.0 | −3.7 |
| LMArena Agent Arena Steering Burden | Tool & computer use › General agents | 21.2 | 14.0 | −3.6 |
| DAYJOB: Tax | Professional › Finance, accounting & tax | 12.8 | 9.2 | −3.6 |
| GDP.pdf (Artificial Analysis, All-pass) | Professional › General knowledge work | 33.4 | 29.8 | −3.6 |
| LMArena Agent Arena Confirmed Success | Tool & computer use › General agents | 19.2 | 12.4 | −3.4 |
| HLAB-Vals-1.1 | Professional › Legal | 19.6 | 16.3 | −3.3 |
| ComplexConstraints (November 2026 rescoring) | Language & writing › Instruction following | 72.9 | 69.6 | −3.3 |
| SimpleQA Verified (Epoch anti-abstention) | Knowledge & reasoning › General knowledge | 75.2 | 78.5 | +3.3 |
| Vals Index v2.2 | Composite indexes | 70.9 | 67.7 | −3.2 |
| Artificial Analysis Healthcare & Medical Index | Health & medicine | 60.1 | 57.0 | −3.1 |
| LiveBench 2.0 (2026-11-06 release) | Composite indexes | 87.4 | 84.9 | −2.6 |
| Agents' Last Exam v1.1 Score | Professional › General knowledge work | 75.8 | 78.3 | +2.5 |
| FrontierMath Tier 4 (v2) - Private | Math | 97.6 | 100.0 | +2.4 |
| AA-Omniscience v1.1 Index | Knowledge & reasoning › General knowledge | 54.8 | 50.0 | −2.4 |
| AutomationBench-AA 1.0.6 | Tool & computer use › Customer service & operations | 79.3 | 77.2 | −2.1 |
| Terminal-Bench 4.0 | Software › Software engineering | 61.6 | 63.1 | +1.5 |
| FrontierMath Tiers 1-3 (v2) - Private | Math | 91.2 | 92.6 | +1.4 |
| Humanity's Last Exam Diamond | Knowledge & reasoning › General knowledge | 71.3 | 72.2 | +0.9 |
| Artificial Analysis Economics Index | Composite indexes | 61.9 | 61.1 | −0.9 |
| BioMysteryBench (Vals Terminus 2) | Science › Life sciences | 80.0 | 80.7 | +0.7 |
| Artificial Analysis Intelligence Index v5.0.1 | Composite indexes | 61.4 | 62.1 | +0.7 |
| GPQA Diamond | Science › Multidisciplinary | 96.3 | 96.8 | +0.6 |
| Terminal-Bench 4.0 (Artificial Analysis) | Software › Software engineering | 63.6 | 64.1 | +0.5 |
| Artificial Analysis Strategy & Ops Index | Professional › General knowledge work | 63.8 | 63.3 | −0.5 |
| Excel Modeling Benchmark v1.1 | Professional › Finance, accounting & tax | 76.5 | 76.0 | −0.4 |
| Artificial Analysis Engineering Index | Science › Engineering | 66.8 | 66.6 | −0.2 |
| LMArena Agent Arena Tool Hallucination | Tool & computer use › Tool calling | 1.0 | 1.2 | +0.1 |
| Furniture Assembly Benchmark | Perception & media › Spatial & visual reasoning | 89.2 | 89.2 | 0.0 |
| OTIS Mock AIME 2024-2025 | Math | 100.0 | 100.0 | 0.0 |
| sudo L7 (October 2026 rescoring) | Software › Software engineering | 38.3 | 38.3 | 0.0 |
Data: Artificial Analysis, Vals AI, Surge AI and 8 othersatlas.village.dev