Atlas

Compare Models

GPT-5.6 Sol vs GPT-6.1 Terra

GPT-6.1 Terra leads by 2.4 capability points (≈95% interval +0.1 to +4.8): a clear lead over GPT-5.6 Sol · P(higher) ≈ 98%

Wins 64 of 95 shared benchmarks (29 lost, 2 tied)

Capability vs. relative cost
GPT-5.6 SolGPT-6.1 TerraClosed 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-5.6 SolvsGPT-6.1 Terracapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins29 – 64 · 95 shared, 2 tied

Comparing matching setups only (6 benchmarks excluded) · suites counted once
← GPT-5.6 SolGPT-6.1 Terra →

GPT-5.6 SolvsGPT-6.1 Terrawins by categoryData Nov 15, 2026Data: benchmarks.bio, Artificial Analysis, OpenAI and 27 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-6.1 Terra minus GPT-5.6 Sol.

HiL-Bench (Human-in-Loop Benchmark)Agents32.368.0+35.7
Terminal-Bench-Science 0.1Science22.457.6+35.2
EBR-bench v4 (card ban, single agent)Games44.878.1+33.3
TissueMicroscopyBench V1 (49 evaluations)Science2.025.9+23.8
Furniture Assembly BenchmarkMultimodal56.779.2+22.5
Legal Research BenchLegal48.126.0−22.1
ExploitBench - Internal Port (June-August 2026)Cybersecurity3.524.5+21.0
ChartographyMultimodal45.063.2+18.2
Humanity's Last Exam DiamondGeneral QA30.447.3+16.9
ALE-BenchCode24102875+16.5
Mystery Game PuzzlesGames58.074.0+16.0
Terminal-Bench 4.0 (Artificial Analysis)Code39.955.6+15.7
LMArena Agent Arena Net Improvement (Steering Burden)Agents4.612.0+13.9
WeirdML v3Science15.028.1+13.1
Tax Agent BenchFinance & Accounting68.055.0−12.9
RareDiseaseBench V1 (35 evaluations)Science13.325.7+12.4
BuseyBench SVGMultimodal6.17.3+11.7
Terminal-Bench 4.0Code37.949.5+11.6
NoCoT-Bench NCRI 15.2General QA109130+11.5
Agents' Last Exam v1.1 (ALE v1.1)Agents30.341.8+11.5
HardGeoBench Country AccuracyMultimodal47.558.8+11.3
ScBench-Long V1 Clean RerunsScience29.618.8−10.8
HardGeoBench Avg DistanceMultimodal14291091+10.4
GBENCH One-shot Coding GScoreGames67.777.6+9.9
FrontierMath Tier 4 (v2) - PrivateMath82.992.7+9.8
MentalHealthBenchSafety46.755.5+8.8
Blueprint-Bench 2Multimodal0.30.4+8.5
LMArena Code Arena WebDevCode16151679+8.3
DistressBenchSafety73.665.9−7.7
NoCoT-Bench NCKI (kspine v3)General QA121110−7.6
MetagenomicsBench V1Science31.338.7+7.3
BioSecBench-Surveillance V1 Clean RerunsScience47.339.9−7.3
FrontierPhysics v0.1Science7.114.3+7.2
LMArena Agent Arena Praise vs ComplaintAgents6.320.0+6.8
TxBench Oligo Discovery V1 (113 evaluations)Science37.230.4−6.8
NeuroscienceBench V1 (91 evaluations)Science15.822.3+6.6
LLM Chess EloGames15501391−6.5
Agents' Last Exam Overall ScoreAgents53.660.0+6.4
GDPval-AA v2.1Professional Work16191476−6.3
TaxCalcBench TY25Finance & Accounting30.024.0−6.0
Toloka Arena (v3)Agents61.055.0−6.0
GameBench 2Games80.185.8+5.8
TxBench Preclinical Pharmacology V1 Clean RerunsScience61.355.7−5.7
LMArena Agent Arena Steering BurdenAgents5.516.6+5.5
BioSecBench-RespiratoryVirus-Surveillance V1 (110 evaluations)Science5.811.2+5.5
SpatialBench V1 Clean RerunsScience68.173.3+5.2
BioSecBench-Structure V1 (58 evaluations)Science20.725.9+5.2
HealthBench Hard (updated implementation)Health & Medicine31.126.0−5.1
HardGeoBench <=100miMultimodal41.346.3+5.0
HardGeoBench <=10miMultimodal30.035.0+5.0
Artificial Analysis Progress Index v1Indexes331345+4.8
HealthBench (updated implementation)Health & Medicine55.651.1−4.5
MacroscopeBenchCode77.882.3+4.5
GPQA DiamondGeneral QA89.994.3+4.4
ChessBench CoherenceGames89.184.8−4.3
CritPt v2 (agentic, avg@5)Science77.181.4+4.3
AA-Briefcase v1.1 EloAgents14791401−4.2
Epoch Capabilities Index (ECI)Indexes163168+4.1
AA-Omniscience IndexIndexes22.029.8+3.9
TxBench-Antibody-Development V1 (60 evaluations)Science18.922.8+3.9
LMArena Agent Arena Confirmed SuccessAgents2.39.7+3.7
Vibe Code Bench v1.1Code80.584.0+3.5
BioSecBench-Function V1 Clean RerunsScience35.539.0+3.5
LiveBench 2.0Agents76.379.7+3.4
TxBench Antibody Discovery V1 (100 evaluations)Science37.340.7+3.3
Vending-Bench 2Games96199107−3.1
SpatialBench-Long V1 Clean RerunsScience37.540.5+3.0
VulcanBench Frontier v4Code87.290.0+2.8
Bug Hunt BenchCode41.438.7−2.7
OncologyBench Strict (25 tasks, October 9 rubric)Health & Medicine18.716.0−2.7
SciCode 1.0.1 (Artificial Analysis)Code57.855.1−2.7
EpiBench V1 Clean RerunsScience25.823.3−2.5
ChessBench AccuracyGames76.478.3+2.0
MMMU ProMultimodal82.784.7+2.0
ChessBench EloGames12181251+2.0
FORTRESS Average Risk Score (ARS)Safety20.118.2+2.0
HealthBench (length-adjusted)Health & Medicine57.055.1−1.9
FrontierMath Tiers 1-3 (v2) - PrivateMath89.190.9+1.8
HealthBench Hard (length-adjusted)Health & Medicine33.131.4−1.7
ScBench V1 Clean RerunsScience59.858.3−1.5
AA-LCR v1.1Search84.082.7−1.3
VoxelBench Text RatingGames22972313+1.1
Artificial Analysis Intelligence Index v4.3.2Indexes47.048.1+1.1
Humanity's Last Exam (Text-Only)General QA47.346.2−1.1
LMArena Agent Arena Bash ReliabilityAgents11.413.4+1.0
GDP.pdf (Artificial Analysis, All-pass)Professional Work27.828.6+0.8
CritPt (avg@5)Science28.628.0−0.6
VariantBench V1 Clean RerunsScience35.334.8−0.6
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine30.030.6+0.6
HealthBench Consensus (updated implementation)Health & Medicine95.394.8−0.5
Prophet Arena Arena Score (Fixed Context)Forecasting84.885.2+0.4
HealthBench ProfessionalHealth & Medicine64.164.2+0.1
AutomationBench-AA 1.0.6Agents60.160.2+0.1
LMArena Agent Arena Tool HallucinationAgents1.01.00.0
OTIS Mock AIME 2024-2025Math100.0100.00.0

GPT-5.6 SolvsGPT-6.1 Terranormalised score marginsData Nov 15, 2026Data: benchmarks.bio, Artificial Analysis, OpenAI and 27 othersVillage 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…