Atlas

Compare Models

GPT-6.1 Sol vs GPT-6.2 Sol

GPT-6.2 Sol leads by 0.6 capability points (≈95% interval −2.1 to +3.4): statistically tied · P(higher) ≈ 67%

Wins 27 of 48 shared benchmarks (21 lost)

Capability vs. relative cost
GPT-6.1 SolGPT-6.2 SolClosed weightsOpen weights

Each line runs through one model’s priced reasoning settings, cheapest to most expensive; grey points are other models’ priced settings in the same range. Results whose reasoning effort the source didn’t report are left out wherever a model has named effort levels. Capability and relative cost are the published estimates (relative cost is cost per task as a multiple of GPT-5 (high)); hover a setting, or use the arrow keys, to set it beside the other model’s same setting.

GPT-6.1 SolvsGPT-6.2 Solcapability vs. relative costData Nov 17, 2026Village LLM Atlasatlas.village.dev

Benchmark wins21 – 27 · 48 shared

Comparing matching setups only (1 benchmark excluded) · suites counted once
← GPT-6.1 SolGPT-6.2 Sol →

GPT-6.1 SolvsGPT-6.2 Solwins by categoryData Nov 17, 2026Data: Artificial Analysis, Vals AI, OpenAI and 1 otherVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-6.2 Sol minus GPT-6.1 Sol.

AA-PersonalityBench Organization indexChat & Writing81.030.8−50.7
AA-PersonalityBench Transparency indexChat & Writing92.447.4−45.4
ExploitBench - Internal Port (June-August 2026)Cybersecurity18.050.5+32.5
AA-PersonalityBench Conviction indexChat & Writing54.385.7+31.7
AA-PersonalityBench Slop indexChat & Writing14.31.0−13.5
Code Migration v1.1 (Vals Index v2.2 subset)Code75.185.3+10.2
SEC-Bench Pro - release-reported cohortCybersecurity78.888.3+9.5
Terminal-Bench-Science 0.1Science58.167.6+9.5
MysteryMechanismScience46.436.9−9.5
ARC-AGI-2General QA94.285.8−8.3
MentalHealthBenchSafety57.950.9−7.0
SecureSWE (OpenAI internal)Cybersecurity49.755.6+5.9
HLAB-Vals-1.1Legal6.312.1+5.8
AA-PersonalityBench Imagination indexChat & Writing61.956.2−5.8
Terminal-Bench Science 0.1 (Vals)Science52.958.6+5.7
Terminal-Bench 4.0Code55.160.6+5.6
GDPval v2 (OpenAI internal)Professional Work41.546.9+5.4
AA-PersonalityBench Warmth indexChat & Writing65.760.5−5.3
AA-PersonalityBench Expressiveness indexChat & Writing50.555.7+5.3
Humanity's Last Exam DiamondGeneral QA58.864.0+5.2
AutomationBench-AA 1.0.6Agents66.671.3+4.7
PatchBench (OpenAI internal)Cybersecurity71.075.6+4.6
LifeVal (GPT-6.2 launch chart)Agents79.583.8+4.3
Terminal-Bench 4.0 (Artificial Analysis)Code56.160.1+4.0
CritPt v2 (agentic, avg@5)Science82.678.6−4.0
FrontierCode 2.0 (Artificial Analysis)Code40.644.4+3.8
AA-AnalystAgent (pass^5)Professional Work50.053.8+3.8
Excel Modeling Benchmark v1.1Finance & Accounting71.174.3+3.2
AA-Briefcase v1.1 EloAgents15551612+3.2
Finance Agent v2.1Finance & Accounting61.858.9−2.9
Artificial Analysis Intelligence Index v5.0.1Indexes54.657.3+2.6
Vals Index v2.2Indexes63.966.1+2.1
Vals Legal IndexIndexes68.766.7−2.0
ARC-AGI-3 (Semi-Private)General QA96.298.2+2.0
GDP.pdf (Artificial Analysis, All-pass)Professional Work32.030.0−2.0
Vals Science IndexIndexes59.657.8−1.9
BioMysteryBench (Vals Terminus 2)Science79.677.8−1.9
AA-PersonalityBench Informativeness indexChat & Writing27.725.9−1.8
HealthBench ProfessionalHealth & Medicine67.265.5−1.7
Legal Research BenchLegal38.537.0−1.4
Vibe Code Bench v1.1Code88.987.6−1.3
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine38.337.2−1.1
AA-Omniscience v1.1 IndexIndexes42.440.4−1.0
ARC-AGI-1 Semi-PrivateGeneral QA98.597.5−1.0
Vals Finance IndexIndexes55.856.7+0.8
Agents' Last Exam v1.1 ScoreAgents72.573.0+0.6
Tax Agent BenchFinance & Accounting62.362.9+0.6
ExploitBench - release-reported exploit progressCybersecurity99.7100.0+0.3

GPT-6.1 SolvsGPT-6.2 Solnormalised score marginsData Nov 17, 2026Data: Artificial Analysis, Vals AI, OpenAI and 1 otherVillage LLM Atlasatlas.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…