Atlas

Compare Models

DeepSeek V4.1 Flash 1017 vs GPT-6.2 Sol

GPT-6.2 Sol leads by 6.2 capability points (≈95% interval +3.5 to +8.9): a clear lead over DeepSeek V4.1 Flash 1017 · P(higher) ≈ 100%

Wins 53 of 68 shared benchmarks (15 lost)

Capability vs. relative cost
DeepSeek V4.1 Flash 1017GPT-6.2 SolClosed weightsOpen weights

Benchmark wins15 – 53 · 68 shared

Comparing matching setups only (32 benchmarks excluded) · suites counted once
← DeepSeek V4.1 Flash 1017GPT-6.2 Sol →
Data: Artificial Analysis, Vals AI, Surge AI and 11 othersatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-6.2 Sol minus DeepSeek V4.1 Flash 1017.

ARC-AGI-3 (Semi-Private)Games5.898.2+92.4
CyberGym-E2E-AA (Artificial Analysis)Software › Cybersecurity76.30.0−76.3
Terminal-Bench-Science 0.1Science › Multidisciplinary12.967.6+54.8
AA-PersonalityBench Organization indexLanguage & writing › Style & personality92.045.7−46.8
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge20.864.0+43.2
RiemannBench (November 2026 rescoring)Math38.480.0+41.6
AA-PersonalityBench Informativeness indexLanguage & writing › Style & personality76.037.7−38.7
AA-PersonalityBench Warmth indexLanguage & writing › Style & personality23.460.1+37.1
AA-PersonalityBench Slop indexLanguage & writing › Style & personality48.915.4−33.9
AA-PersonalityBench Transparency indexLanguage & writing › Style & personality88.856.9−32.3
DAYJOB: Law (November 2026 release)Professional › Legal2.832.7+29.9
AA-PersonalityBench Imagination indexLanguage & writing › Style & personality31.360.1+29.0
Terminal-Bench 4.0Software › Software engineering31.860.6+28.8
Humanity's Sixth SensePerception & media › Spatial & visual reasoning30.156.8+26.7
DeepsecBench-AA (Artificial Analysis)Software › Cybersecurity24.951.6+26.6
Terminal-Bench 4.0 (Artificial Analysis)Software › Software engineering33.860.1+26.3
CritPt v2 (agentic, avg@5)Science › Physical sciences52.377.7+25.4
Artificial Analysis Intelligence Index v5.0.1Composite indexes32.657.3+24.7
Artificial Analysis Engineering IndexScience › Engineering37.661.2+23.6
VIALS-AA (pass^3)Science › Life sciences40.462.7+22.4
EnterpriseBench: CoreCraftTool & computer use › Customer service & operations58.536.4−22.1
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents0.811.3+20.5
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax12.532.8+20.3
DAYJOB: AccountingProfessional › Finance, accounting & tax7.627.6+20.0
AA-Omniscience v1.1 IndexKnowledge & reasoning › General knowledge1.540.4+19.4
Ramp SWE-Bench v2 (Bash harness)Software › Software engineering53.472.4+19.0
Artificial Analysis Economics IndexComposite indexes38.557.4+18.9
DeepsecBench — Revalidated V1 (233 issues)Software › Cybersecurity30.047.3+17.4
Artificial Analysis Legal IndexProfessional › Legal37.755.0+17.3
Code Migration v1.1 (Vals Index v2.2 subset)Software › Software engineering68.585.3+16.8
AA-PersonalityBench Conviction indexLanguage & writing › Style & personality95.279.2−16.1
FrontierCode 2.0 (Artificial Analysis)Software › Software engineering28.544.4+15.9
VIALS-AA (average of 3)Science › Life sciences53.268.9+15.7
Artificial Analysis Finance & Accounting IndexProfessional › Finance, accounting & tax42.257.2+15.1
HealthBench Professional (Length-Adjusted)Health & medicine51.765.9+14.2
GDP.pdf (Artificial Analysis, All-pass)Professional › General knowledge work17.230.0+12.8
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents-1.523.3+12.4
Tax Agent BenchProfessional › Finance, accounting & tax74.962.9−12.0
Artificial Analysis Healthcare & Medical IndexHealth & medicine41.353.3+12.0
Bug Hunt BenchSoftware › Software engineering40.252.1+11.9
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine6.017.2+11.2
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine26.737.2+10.6
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations24.034.5+10.5
Artificial Analysis Strategy & Ops IndexProfessional › General knowledge work49.259.4+10.2
LiveBench 2.0 (2026-11-06 release)Composite indexes70.380.3+10.0
Legal Research BenchProfessional › Legal45.237.0−8.2
VIALS-AA (pass@3)Science › Life sciences66.574.5+8.1
ARC-AGI-2Knowledge & reasoning › Abstract reasoning78.185.8+7.8
CWE-Bench-AA (Artificial Analysis)Software › Cybersecurity69.261.7−7.5
AA-Briefcase v1.1 EloProfessional › General knowledge work14161541+6.8
Agents' Last Exam v1.1 ScoreProfessional › General knowledge work66.673.0+6.4
LMArena Agent Arena Steering BurdenTool & computer use › General agents2.313.4+5.5
Vals Legal IndexProfessional › Legal71.966.7−5.2
Excel Modeling Benchmark v1.1Professional › Finance, accounting & tax69.274.3+5.1
AA-PersonalityBench Expressiveness indexLanguage & writing › Style & personality50.555.3+4.8
Vals Index v2.2Composite indexes61.366.1+4.8
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents2.210.9+4.3
Vals Finance IndexProfessional › Finance, accounting & tax60.956.7−4.2
Finance Agent v2.1Professional › Finance, accounting & tax62.958.9−4.0
GDP.xlsxProfessional › General knowledge work18.422.3+3.9
DAYJOB: TaxProfessional › Finance, accounting & tax1.24.8+3.6
ARC-AGI-1 Semi-PrivateKnowledge & reasoning › Abstract reasoning94.597.5+3.0
AutomationBench-AA 1.0.6Tool & computer use › Customer service & operations68.571.3+2.8
HLAB-Vals-1.1Professional › Legal14.212.1−2.1
DualEntry Accounting AI BenchmarkProfessional › Finance, accounting & tax82.284.2+2.0
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents1.14.3+1.6
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling-0.11.2+0.6
Vibe Code Bench v1.1Software › Frontend & design87.787.6−0.1
Data: Artificial Analysis, Vals AI, Surge AI and 11 othersatlas.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…