Atlas

Compare Models

Gemini 4 Carbon vs Gemini 4 Hydrogen

Gemini 4 Carbon leads by 2.3 capability points (≈95% interval −0.7 to +5.3): statistically tied · P(higher) ≈ 94%

Wins 46 of 52 shared benchmarks (6 lost)

Capability vs. relative cost
Gemini 4 CarbonGemini 4 HydrogenClosed 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 CarbonvsGemini 4 Hydrogencapability vs. relative costData Nov 17, 2026Village LLM Atlasatlas.village.dev

Benchmark wins46 – 6 · 52 shared

Comparing matching setups only (1 benchmark excluded) · suites counted once
← Gemini 4 CarbonGemini 4 Hydrogen →

Gemini 4 CarbonvsGemini 4 Hydrogenwins by categoryData Nov 17, 2026Data: Artificial Analysis, Vals AI, Arena and 11 othersVillage LLM Atlasatlas.village.dev

Margins

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

Terminal-Bench-Science 0.1Science62.415.2−47.1
Terminal-Bench 4.0 (Artificial Analysis)Code60.629.3−31.3
Bug Hunt BenchCode54.627.0−27.6
Artificial Analysis Intelligence Index v5.0.1Indexes60.041.2−18.8
FrontierSWE V2 Worst@5Code56.245.3−10.8
Humanity's Last Exam DiamondGeneral QA69.158.8−10.3
BuseyBench SVGMultimodal7.96.9−10.0
Terminal-Bench 4.0Code63.154.0−9.1
LMArena Code Arena WebDevCode17681704−8.5
TxBench Antibody Discovery V1 (100 evaluations)Science65.057.7−7.3
HLAB-Vals-1.1Legal17.511.3−6.2
AA-Briefcase v1.1 EloAgents16111497−6.2
LMArena Text Arena EloChat & Writing15261490−6.2
CursorBench 4.0Code60.454.3−6.1
ChartographyMultimodal72.566.6−5.9
Agents' Last Exam v1.1 ScoreAgents77.471.6−5.8
FrontierSWE V2 Best@5Code66.761.2−5.5
Artificial Analysis Legal IndexIndexes62.056.7−5.2
Artificial Analysis Engineering IndexIndexes65.360.1−5.2
FrontierCode 2.0 - Pass RateCode46.942.2−4.7
GradeBenchProfessional Work56.651.9−4.7
TxBench Preclinical Pharmacology V1 Clean RerunsScience66.061.3−4.7
Artificial Analysis Finance & Accounting IndexIndexes62.658.1−4.6
Artificial Analysis Economics IndexIndexes59.655.0−4.6
FrontierCode 2.0 - ScoreCode42.638.2−4.4
CritPt v2 (agentic, avg@5)Science82.378.0−4.3
OSWorld 2.1 Offline, 500-step Partial ScoreAgents81.577.6−3.9
AA-Omniscience v1.1 IndexIndexes53.546.3−3.6
Artificial Analysis Healthcare & Medical IndexIndexes56.853.3−3.4
AutomationBench-AA 1.0.6Agents77.874.4−3.4
LMArena Agent Arena Praise vs ComplaintAgents40.333.8−3.3
Artificial Analysis Strategy & Ops IndexIndexes60.957.7−3.2
GDP.pdf (Artificial Analysis, All-pass)Professional Work29.632.8+3.2
Artificial Analysis Coding Agent Index v2.1Indexes55.552.3−3.1
LMArena Agent Arena Net Improvement (Steering Burden)Agents16.414.9−2.9
Code Migration v1.1 (Vals Index v2.2 subset)Code80.182.7+2.6
Vals Legal IndexIndexes75.973.6−2.3
GBENCH One-shot Coding GScoreGames69.066.8−2.2
Finance Agent v2.1Finance & Accounting67.265.1−2.0
Tax Agent BenchFinance & Accounting76.478.4+2.0
Legal Research BenchLegal55.853.8−1.9
AutomationBench 1.0.6 (Zapier)Agents53.051.1−1.8
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine19.421.1+1.7
TxBench Oligo Discovery V1 (113 evaluations)Science55.854.3−1.5
Vibe Code Bench v1.1Code89.187.8−1.3
Vals Finance IndexIndexes67.066.1−0.9
Excel Modeling Benchmark v1.1Finance & Accounting76.575.8−0.7
LMArena Agent Arena Bash ReliabilityAgents-5.9-4.6+0.7
TxBench-Antibody-Development V1 (60 evaluations)Science61.160.6−0.5
LMArena Agent Arena Confirmed SuccessAgents17.916.8−0.5
LMArena Agent Arena Steering BurdenAgents18.118.4+0.1
LMArena Agent Arena Tool HallucinationAgents0.60.5−0.1

Gemini 4 CarbonvsGemini 4 Hydrogennormalised score marginsData Nov 17, 2026Data: Artificial Analysis, Vals AI, Arena and 11 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…