Atlas

Compare Models

Gemini 4 Carbon vs GPT-6 Sol

Gemini 4 Carbon leads by 4.2 capability points (≈95% interval +1.5 to +6.8): a clear lead over GPT-6 Sol · P(higher) ≈ 100%

Wins 38 of 43 shared benchmarks (4 lost, 1 tied)

Capability vs. relative cost
Gemini 4 CarbonGPT-6 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.

Gemini 4 CarbonvsGPT-6 Solcapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins38 – 4 · 43 shared, 1 tied

Comparing matching setups only (35 benchmarks excluded) · suites counted once
← Gemini 4 CarbonGPT-6 Sol →

Gemini 4 CarbonvsGPT-6 Solwins by categoryData Nov 15, 2026Data: Artificial Analysis, Arena, Mercor and 9 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-6 Sol minus Gemini 4 Carbon.

ARC-AGI-3 (Semi-Private)General QA99.14.9−94.2
Terminal-Bench-Science 0.1Science62.48.6−53.8
Terminal-Bench 4.0 (Artificial Analysis)Code60.618.7−41.9
Legal Research BenchLegal55.828.8−26.9
Humanity's Last Exam DiamondGeneral QA69.145.2−23.9
Tax Agent BenchFinance & Accounting76.453.0−23.4
APEX-Agents 1.1Professional Work90.468.7−21.7
Blueprint-Bench 2Multimodal0.60.4−21.7
TxBench Antibody Discovery V1 (100 evaluations)Science65.043.3−21.7
AutomationBench 1.0.6 (Zapier)Agents53.033.2−19.8
LMArena Agent Arena Net Improvement (Steering Burden)Agents17.06.9−19.0
ChartographyMultimodal72.553.6−18.9
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal23.04.1−18.9
AA-Briefcase v1.1 EloAgents16111266−18.9
AutomationBench-AA 1.0.6Agents77.861.7−16.1
AA-Omniscience v1.1 IndexIndexes53.521.4−16.1
Artificial Analysis Intelligence Index v5.0.1Indexes60.044.2−15.9
LMArena Agent Arena Praise vs ComplaintAgents41.010.0−15.5
RiemannBenchMath80.064.8−15.2
GDPval-AA v2.1Professional Work17331402−14.6
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine19.433.9+14.4
Terminal-Bench 4.0Code63.149.4−13.7
τ³-Banking (Mercor run, bm25_grep retrieval)Agents55.743.0−12.7
TxBench Oligo Discovery V1 (113 evaluations)Science55.843.1−12.7
LMArena Text Arena EloChat & Writing15261457−12.1
APEX-Accounting 1.1Finance & Accounting70.759.7−11.0
BrokenArXiv 08/2026 (Native Harnesses)Math88.178.6−9.5
LMArena Agent Arena Bash ReliabilityAgents-5.210.9+8.1
LMArena Agent Arena Confirmed SuccessAgents18.44.5−6.9
MMMU ProMultimodal89.282.5−6.7
Artificial Analysis Progress Index v1Indexes351334−5.9
FrontierCode 2.0 (Artificial Analysis)Code40.635.6−5.0
ARC-AGI-2General QA94.289.6−4.6
LMArena Agent Arena Steering BurdenAgents19.110.1−4.5
GDP.pdf (Artificial Analysis, All-pass)Professional Work29.625.2−4.4
AA-LCR v1.1Search79.383.7+4.3
BuseyBench SVGMultimodal7.97.5−3.3
ARC-AGI-1 Semi-PrivateGeneral QA98.595.5−3.0
TxBench Preclinical Pharmacology V1 Clean RerunsScience66.063.0−3.0
CritPt v2 (agentic, avg@5)Science82.380.9−1.4
Vibe Code Bench v1.1Code89.187.8−1.2
LMArena Agent Arena Tool HallucinationAgents0.51.0+0.2
ArXivMath 08/2026 (Native Harnesses)Math93.093.00.0

Gemini 4 CarbonvsGPT-6 Solnormalised score marginsData Nov 15, 2026Data: Artificial Analysis, Arena, Mercor and 9 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…