Atlas

Compare Models

Claude Opus 5.5 vs GPT-6.2 Astra

GPT-6.2 Astra leads by 0.9 capability points (≈95% interval −1.9 to +3.7): statistically tied · P(higher) ≈ 74%

Wins 41 of 70 shared benchmarks (26 lost, 3 tied)

Capability vs. relative cost
Claude Opus 5.5GPT-6.2 AstraClosed weightsOpen weights

Benchmark wins26 – 41 · 70 shared, 3 tied

Comparing matching setups only (63 benchmarks excluded) · suites counted once
← Claude Opus 5.5GPT-6.2 Astra →
Data: Artificial Analysis, Surge AI, Epoch AI and 18 othersatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-6.2 Astra minus Claude Opus 5.5.

AA-PersonalityBench Informativeness indexLanguage & writing › Style & personality90.531.3−59.7
AA-PersonalityBench Organization indexLanguage & writing › Style & personality79.121.8−57.9
AA-PersonalityBench Transparency indexLanguage & writing › Style & personality90.540.9−50.1
TxBench Preclinical Pharmacology V1 Clean RerunsScience › Life sciences26.368.0+41.7
SpireClimbGames5.841.3+35.5
DeepsecBench — Revalidated V1 (233 issues)Software › Cybersecurity26.661.2+34.7
AA-PersonalityBench Imagination indexLanguage & writing › Style & personality94.363.3−31.3
EnterpriseBench: CoreCraftTool & computer use › Customer service & operations74.443.6−30.8
MirrorCode (ML, +Private, 2L)Software › Software engineering95.667.8−27.8
AA-PersonalityBench Expressiveness indexLanguage & writing › Style & personality90.566.5−24.3
Mystery Game PuzzlesGames71.095.0+24.0
EBR-bench v4 (card ban, single agent)Games71.492.4+21.0
Humanity's Sixth SensePerception & media › Spatial & visual reasoning44.663.2+18.6
DAYJOB: Law (November 2026 release)Professional › Legal23.741.4+17.7
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge54.672.2+17.6
SWE Atlas v1.1 - OverallSoftware › Software engineering77.560.5−17.0
AA-PersonalityBench Warmth indexLanguage & writing › Style & personality84.868.1−16.9
Terminal-Bench-Science 0.1Science › Multidisciplinary61.978.6+16.7
AA-3D v1.0 EloSoftware › Frontend & design12021318+15.7
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine48.334.4−13.9
AA-PersonalityBench Slop indexLanguage & writing › Style & personality20.07.4−12.8
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & medicine27.839.4+11.7
SimpleBenchKnowledge & reasoning › General knowledge88.477.3−11.1
Chess Puzzles (task version 2.0)Games67.078.0+11.0
Blueprint-Bench 2Perception & media › Spatial & visual reasoning0.50.6+10.2
AA-Briefcase v1.1 EloProfessional › General knowledge work18001644−8.6
sudo L7 (October 2026 rescoring)Software › Software engineering46.138.3−7.8
DAYJOB: AccountingProfessional › Finance, accounting & tax34.442.0+7.6
FrontierMath ErdősMath2.910.3+7.4
HLAB-Vals-1.1Professional › Legal9.216.3+7.1
GDPval v2 (OpenAI internal)Professional › General knowledge work43.550.4+6.9
LMArena Agent Arena Net Improvement (Steering Burden)Tool & computer use › General agents18.615.2−6.6
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations52.345.8−6.5
SimpleQA Verified (Epoch anti-abstention)Knowledge & reasoning › General knowledge72.278.5+6.3
AA-AnalystAgent (pass^5)Professional › General knowledge work56.362.5+6.3
Furniture Assembly BenchmarkPerception & media › Spatial & visual reasoning83.389.2+5.8
BigFinanceBench private holdout (final-answer accuracy)Professional › Finance, accounting & tax48.254.0+5.8
ComplexConstraints (November 2026 rescoring)Language & writing › Instruction following64.469.6+5.2
FrontierCode 2.0 (Artificial Analysis)Software › Software engineering43.849.0+5.2
FrontierMath Tier 4 (v2) - PrivateMath95.0100.0+5.0
Terminal-Bench 4.0 (Artificial Analysis)Software › Software engineering59.664.1+4.5
Professional Reasoning Benchmark - LegalProfessional › Legal47.643.1−4.5
DualEntry Accounting AI BenchmarkProfessional › Finance, accounting & tax82.278.2−4.0
RiemannBench (November 2026 rescoring)Math80.076.0−4.0
Ramp SWE-Bench v2 (Bash harness)Software › Software engineering71.275.0+3.8
LMArena Agent Arena Confirmed SuccessTool & computer use › General agents19.212.4−3.4
Epoch Capabilities Index (ECI)Composite indexes169173+3.3
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax42.545.8+3.3
LMArena Agent Arena Bash ReliabilityTool & computer use › General agents14.07.7−3.2
MMMU ProKnowledge & reasoning › General knowledge86.689.5+2.9
ARC-AGI-2Knowledge & reasoning › Abstract reasoning93.395.8+2.5
Dual-Arm Sim (task completion)Tool & computer use › Robotics0.02.5+2.5
CritPt v2 (agentic, avg@5)Science › Physical sciences81.783.7+2.0
TaxCalcBench TY25Professional › Finance, accounting & tax42.044.0+2.0
LMArena Agent Arena Steering BurdenTool & computer use › General agents18.014.0−2.0
AA-PersonalityBench Conviction indexLanguage & writing › Style & personality69.571.3+1.7
LiveBench 2.0 (2026-11-06 release)Composite indexes83.484.9+1.4
FrontierMath Tiers 1-3 (v2) - PrivateMath91.292.6+1.4
Gray Swan IPI Attack Success (k=15)Safety & alignment1.02.4−1.4
Chartography (November 2026 rescoring)Perception & media › Documents & charts89.690.7+1.1
LLM Debate BenchmarkLanguage & writing › Language & wordplay16801673−1.0
GDP.pdf (Artificial Analysis, All-pass)Professional › General knowledge work28.829.8+1.0
LMArena Agent Arena Praise vs ComplaintTool & computer use › General agents38.637.0−0.8
GDP.xlsxProfessional › General knowledge work30.329.7−0.6
GDP.pdf (November 2026 rescoring)Professional › General knowledge work48.849.2+0.4
Excel Modeling Benchmark v1.1Professional › Finance, accounting & tax75.876.0+0.2
LMArena Agent Arena Tool HallucinationTool & computer use › Tool calling1.01.2+0.1
ARC-AGI-1 Semi-PrivateKnowledge & reasoning › Abstract reasoning98.598.50.0
DAYJOB: TaxProfessional › Finance, accounting & tax9.29.20.0
OTIS Mock AIME 2024-2025Math100.0100.00.0
Data: Artificial Analysis, Surge AI, Epoch AI and 18 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…