Atlas

Compare Models

GPT-5.6 Sol vs GPT-6.2 Astra

GPT-6.2 Astra leads by 5.9 capability points (≈95% interval +3.4 to +8.5): a clear lead over GPT-5.6 Sol · P(higher) ≈ 100%

Wins 56 of 64 shared benchmarks (6 lost, 2 tied)

Capability vs. relative cost
GPT-5.6 SolGPT-6.2 AstraClosed weightsOpen weights

Benchmark wins6 – 56 · 64 shared, 2 tied

Comparing matching setups only (7 benchmarks excluded) · suites counted once
← GPT-5.6 SolGPT-6.2 Astra →
Data: Artificial Analysis, Epoch AI, Vals AI and 17 othersatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-6.2 Astra minus GPT-5.6 Sol.

ARC-AGI-3 (Semi-Private)Games7.870.3+62.5
Terminal-Bench-Science 0.1Science › Multidisciplinary22.478.1+55.7
Terminal-Bench Science 0.1 (Vals)Science › Multidisciplinary20.071.4+51.4
MirrorCode (ML, +Private, 2L)Software › Software engineering20.067.8+47.8
EBR-bench v4 (card ban, single agent)Games44.892.4+47.6
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge30.472.2+41.8
Mystery Game PuzzlesGames58.095.0+37.0
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax11.845.8+34.0
Humanity's Sixth SensePerception & media › Spatial & visual reasoning30.063.2+33.2
Furniture Assembly BenchmarkPerception & media › Spatial & visual reasoning56.789.2+32.5
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine4.434.4+30.0
Blueprint-Bench 2Perception & media › Spatial & visual reasoning0.30.6+27.8
Vals Science IndexScience › Multidisciplinary41.568.4+27.0
DeepsecBench — Revalidated V1 (233 issues)Software › Cybersecurity35.361.2+25.9
Terminal-Bench 4.0Software › Software engineering37.963.1+25.3
FrontierCUA v1 pass@1Tool & computer use › Computer & browser use7.532.5+25.0
MedChemBenchScience › Life sciences59.884.5+24.7
Gray Swan IPI Attack Success (k=15)Safety & alignment27.02.4+24.6
Terminal-Bench 4.0 (Artificial Analysis)Software › Software engineering39.964.1+24.2
AutomationBench 1.0.6 (Zapier)Tool & computer use › Customer service & operations28.852.4+23.6
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents3.715.2+22.5
LMArena Code Arena WebDevSoftware › Frontend & design16151781+22.0
ExploitBench - release-reported exploit progressSoftware › Cybersecurity78.5100.0+21.5
MysteryMechanismMath33.353.2+19.8
Chess Puzzles (task version 2.0)Games59.078.0+19.0
Bug Hunt BenchSoftware › Software engineering41.459.7+18.3
AutomationBench-AA 1.0.6Tool & computer use › Customer service & operations60.177.2+17.2
FrontierMath Tier 4 (v2) - PrivateMath82.9100.0+17.1
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents3.837.0+16.6
Agents' Last Exam v1.1 (ALE v1.1)Professional › General knowledge work30.346.7+16.4
AA-AnalystAgent (pass^5)Professional › General knowledge work47.562.5+15.0
EnterpriseOps-Gym-AATool & computer use › Customer service & operations42.929.9−13.0
Chartography (November 2026 rescoring)Perception & media › Documents & charts78.690.7+12.1
TaxCalcBench TY25Professional › Finance, accounting & tax32.044.0+12.0
FrontierMath ErdősMath0.010.3+10.3
Tax Agent BenchProfessional › Finance, accounting & tax68.057.9−10.1
BigFinanceBench private holdout (final-answer accuracy)Professional › Finance, accounting & tax44.254.0+9.8
BioMysteryBench (Vals Terminus 2)Science › Life sciences71.180.7+9.6
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine30.039.4+9.4
Vibe Code Bench v1.1Software › Frontend & design80.589.9+9.4
AA-Briefcase v1.1 EloProfessional › General knowledge work14791644+9.0
DrugDiscoveryBenchScience › Life sciences56.565.5+9.0
Epoch Capabilities Index (ECI)Composite indexes163173+7.9
SEC-Bench Pro - release-reported cohortSoftware › Cybersecurity79.186.8+7.7
RiemannBench (November 2026 rescoring)Math83.276.0−7.2
LLM Debate BenchmarkLanguage & writing › Language & wordplay16251673+6.9
MMMU ProKnowledge & reasoning › General knowledge82.789.5+6.8
TxBench Preclinical Pharmacology V1 Clean RerunsScience › Life sciences61.368.0+6.7
MentalHealthBenchHealth & medicine46.753.3+6.6
CritPt v2 (agentic, avg@5)Science › Physical sciences77.183.7+6.6
Legal Research BenchProfessional › Legal48.142.8−5.3
Vals RSI Index v1.1 (four tasks)Software › Data & ML23.929.0+5.1
LMArena Agent Arena Steering BurdenTool & computer use › General agents3.814.0+5.1
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents2.512.4+5.0
FrontierMath Tiers 1-3 (v2) - PrivateMath89.192.6+3.5
ARC-AGI-2Knowledge & reasoning › Abstract reasoning92.595.8+3.3
EnterpriseBench: CoreCraftTool & computer use › Customer service & operations46.243.6−2.6
GPQA DiamondScience › Multidisciplinary94.697.0+2.4
GDP.pdf (Artificial Analysis, All-pass)Professional › General knowledge work27.829.8+2.0
HealthBench ProfessionalHealth & medicine64.166.0+1.9
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents10.57.7−1.4
ARC-AGI-1 Semi-PrivateKnowledge & reasoning › Abstract reasoning97.598.5+1.0
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling1.21.20.0
OTIS Mock AIME 2024-2025Math100.0100.00.0
Data: Artificial Analysis, Epoch AI, Vals AI and 17 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…