Atlas

Compare Models

Gemini 4 Argon vs Gemini 4 Carbon

Gemini 4 Carbon leads by 2.0 capability points (≈95% interval −0.7 to +4.7): statistically tied · P(higher) ≈ 93%

Wins 67 of 79 shared benchmarks (11 lost, 1 tied)

Capability vs. relative cost
Gemini 4 ArgonGemini 4 CarbonClosed 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 ArgonvsGemini 4 Carboncapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins11 – 67 · 79 shared, 1 tied

Comparing matching setups only · suites counted once
← Gemini 4 ArgonGemini 4 Carbon →

Gemini 4 ArgonvsGemini 4 Carbonwins by categoryData Nov 15, 2026Data: Artificial Analysis, Vals AI, Arena and 18 othersVillage LLM Atlasatlas.village.dev

Margins

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

BrokenArXiv 10/2026 (Native Harnesses)Math39.882.7+42.9
BrokenArXiv 08/2026 (Native Harnesses)Math54.288.1+33.9
BrokenArXiv 09/2026 (Native Harnesses)Math49.782.4+32.7
TxBench-Antibody-Development V1 (60 evaluations)Science31.161.1+30.0
Humanity's Last Exam DiamondGeneral QA42.569.1+26.6
ArXivLean 09/2026Math39.662.3+22.6
FrontierCode 2.0 - Pass RateCode30.646.9+16.3
Bug Hunt BenchCode38.754.6+15.9
FrontierCode 2.0 - ScoreCode27.642.6+15.0
Terminal-Bench-Science 0.1Science48.662.4+13.8
ExploitGym - release-reported percentage, cohort unspecifiedCybersecurity42.054.8+12.8
GraphWalks BFS: 256K to 1M AggregateGeneral QA84.296.2+12.0
LMArena Agent Arena Net Improvement (Steering Burden)Agents10.717.0+11.8
FrontierSWE V2 Worst@5Code44.556.2+11.7
LMArena Code Arena WebDevCode16821768+11.3
ArXivMath 10/2026 (Native Harnesses)Math83.093.6+10.6
ArXivMath 08/2026 (Native Harnesses)Math82.593.0+10.5
Artificial Analysis Coding Agent Index v2.1Indexes45.555.5+10.0
Artificial Analysis Intelligence Index v5.0.1Indexes50.460.0+9.6
HLAB-Vals-1.1Legal27.117.5−9.6
CritPt v2 (agentic, avg@5)Science72.982.3+9.4
OpenWorldEngineer (Mean@8)Code24.132.8+8.7
ArXivMath 09/2026 (Native Harnesses)Math87.796.2+8.5
LMArena Agent Arena Praise vs ComplaintAgents24.141.0+8.5
Agents' Last Exam v1.1 ScoreAgents69.077.4+8.4
CursorBench 4.0Code52.960.4+7.5
AA-Briefcase v1.1 EloAgents14851611+6.9
OSWorld 2.1 Offline, 500-step Partial ScoreAgents74.781.5+6.8
Artificial Analysis Coding Agent Index v2.0Indexes53.259.4+6.2
Code Migration v1.1 (Vals Index v2.2 subset)Code74.480.1+5.6
Terminal-Bench 4.0 (Artificial Analysis)Code55.160.6+5.6
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine13.919.4+5.6
Terminal-Bench 4.0Code57.663.1+5.6
DAXBench v2Code88.393.6+5.3
SEC-Bench Pro - release-reported cohortCybersecurity83.488.2+4.8
TxBench Oligo Discovery V1 (113 evaluations)Science51.055.8+4.7
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal27.423.0−4.4
GDPval-AA v2.1Professional Work16341733+4.4
VIALS-AA (pass^3)Multimodal59.664.0+4.3
Blueprint-Bench 2Multimodal0.50.6+4.2
GDP.pdf (Artificial Analysis, All-pass)Professional Work25.629.6+4.0
RiemannBenchMath76.080.0+4.0
AA-Omniscience v1.1 IndexIndexes46.153.5+3.7
GDPxProfessional Work7.310.9+3.5
Artificial Analysis Progress Index v1Indexes342351+3.2
MKA (Artificial Analysis)Professional Work51.354.3+3.1
BuseyBench SVGMultimodal7.67.9+3.0
APEX-Agents 1.1Professional Work87.490.4+3.0
CWE-bench v1 Programmatic Pass@1Cybersecurity68.071.0+3.0
LMArena Agent Arena Steering BurdenAgents13.219.1+3.0
ARC-AGI-2General QA91.394.2+2.9
AutomationBench 1.0.6 (Zapier)Agents50.153.0+2.9
Vals RSI Index v1.1 (four tasks)Agents30.627.7−2.8
Tax Agent BenchFinance & Accounting79.076.4−2.6
LMArena Agent Arena Confirmed SuccessAgents13.218.4+2.6
τ³-Banking (Mercor run, bm25_grep retrieval)Agents58.155.7−2.4
FrontierSWE V2 Best@5Code64.366.7+2.4
FrontierBioMLScience36.839.0+2.1
VIALS-AA (pass@3)Multimodal72.770.8−1.9
ARC-AGI-3 (Semi-Private)General QA97.399.1+1.7
Excel Modeling Benchmark v1.1Finance & Accounting74.876.5+1.6
TxBench Antibody Discovery V1 (100 evaluations)Science63.765.0+1.3
TxBench Preclinical Pharmacology V1 Clean RerunsScience64.766.0+1.3
APEX-1 (Mercor, mean score)Professional Work81.180.0−1.1
Vals Finance IndexIndexes66.067.0+1.0
VIALS-AA (average of 3)Multimodal66.367.3+1.0
Legal Research BenchLegal54.855.8+1.0
MMMU ProMultimodal88.389.2+0.9
ChartographyMultimodal71.672.5+0.9
Vals Legal IndexIndexes76.875.9−0.8
APEX-Accounting 1.1Finance & Accounting71.570.7−0.8
Gray Swan IPI Attack Success (k=15)Safety0.70.0+0.7
LMArena Text Arena EloChat & Writing15221526+0.6
Finance Agent v2.1Finance & Accounting67.767.2−0.5
AA-LCR v1.1Search79.779.3−0.3
AutomationBench-AA 1.0.6Agents77.577.8+0.3
LMArena Agent Arena Bash ReliabilityAgents-5.4-5.2+0.1
LMArena Agent Arena Tool HallucinationAgents0.50.50.0
ARC-AGI-1 Semi-PrivateGeneral QA98.598.50.0

Gemini 4 ArgonvsGemini 4 Carbonnormalised score marginsData Nov 15, 2026Data: Artificial Analysis, Vals AI, Arena and 18 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…