Atlas

Compare Models

DeepSeek V4.1 Flash 1017 vs Gemini 4 Carbon

Gemini 4 Carbon leads by 7.0 capability points (≈95% interval +4.3 to +9.7): a clear lead over DeepSeek V4.1 Flash 1017 · P(higher) ≈ 100%

Wins 61 of 64 shared benchmarks (3 lost)

Capability vs. relative cost
DeepSeek V4.1 Flash 1017Gemini 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.

DeepSeek V4.1 Flash 1017vsGemini 4 Carboncapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins3 – 61 · 64 shared

Comparing matching setups only (3 benchmarks excluded) · suites counted once
← DeepSeek V4.1 Flash 1017Gemini 4 Carbon →

DeepSeek V4.1 Flash 1017vsGemini 4 Carbonwins by categoryData Nov 15, 2026Data: Artificial Analysis, Vals AI, Arena and 12 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 DeepSeek V4.1 Flash 1017.

ARC-AGI-3 (Semi-Private)General QA5.899.1+93.3
ArXivMath 10/2026 (Native Harnesses)Math38.393.6+55.3
ArXivMath 09/2026 (Native Harnesses)Math45.396.2+51.0
Terminal-Bench-Science 0.1Science12.962.4+49.5
ArXivMath 08/2026 (Native Harnesses)Math43.993.0+49.1
Humanity's Last Exam DiamondGeneral QA20.869.1+48.3
RiemannBench (November 2026 rescoring)Math38.480.0+41.6
BrokenArXiv 08/2026 (Native Harnesses)Math46.788.1+41.4
BuseyBench SVGMultimodal4.37.9+35.1
BrokenArXiv 10/2026 (Native Harnesses)Math49.782.7+33.0
BrokenArXiv 09/2026 (Native Harnesses)Math49.782.4+32.7
Terminal-Bench 4.0Code31.863.1+31.3
CritPt v2 (agentic, avg@5)Science52.382.3+30.0
Artificial Analysis Intelligence Index v5.0Indexes28.557.7+29.2
Terminal-Bench 4.0 (Artificial Analysis)Code33.862.6+28.8
LMArena Agent Arena Net Improvement (Steering Burden)Agents2.017.0+28.3
Artificial Analysis Engineering IndexIndexes37.665.3+27.7
Artificial Analysis Intelligence Index v5.0.1Indexes32.660.0+27.4
AA-Omniscience v1.1 IndexIndexes1.553.5+26.0
Artificial Analysis Legal IndexIndexes37.762.0+24.2
FrontierCode 2.0 - Pass RateCode23.046.9+23.9
VIALS-AA (pass^3)Multimodal40.464.0+23.6
FrontierCode 2.0 - ScoreCode20.242.6+22.4
Artificial Analysis Economics IndexIndexes38.559.6+21.1
Artificial Analysis Finance & Accounting IndexIndexes42.262.6+20.5
LMArena Agent Arena Praise vs ComplaintAgents1.241.0+19.9
Artificial Analysis Coding Agent Index v2.1Indexes36.855.5+18.6
ARC-AGI-2General QA78.194.2+16.1
DAXBench v2Code77.793.6+16.0
MKA (Artificial Analysis)Professional Work38.554.3+15.8
Artificial Analysis Healthcare & Medical IndexIndexes41.356.8+15.5
τ³-Banking (Mercor run, bm25_grep retrieval)Agents41.655.7+14.1
VIALS-AA (average of 3)Multimodal53.267.3+14.1
Bug Hunt BenchCode40.754.6+13.9
LMArena Code Arena WebDevCode16641768+13.7
CursorBench 4.0Code47.760.4+12.7
GDP.pdf (Artificial Analysis, All-pass)Professional Work17.229.6+12.4
Artificial Analysis Strategy & Ops IndexIndexes49.260.9+11.7
Code Migration v1.1 (Vals Index v2.2 subset)Code68.580.1+11.6
APEX-Accounting 1.1Finance & Accounting59.370.7+11.4
APEX-Agents 1.1Professional Work79.490.4+11.0
Agents' Last Exam v1.1 ScoreAgents66.677.4+10.8
AA-Briefcase v1.1 EloAgents14161611+10.7
Legal Research BenchLegal45.255.8+10.6
MMMU ProMultimodal79.789.2+9.5
Artificial Analysis Progress Index v1Indexes324351+9.5
AutomationBench-AA 1.0.6Agents68.577.8+9.3
LMArena Agent Arena Confirmed SuccessAgents2.318.4+8.0
LMArena Agent Arena Steering BurdenAgents4.019.1+7.5
Excel Modeling Benchmark v1.1Finance & Accounting69.276.5+7.3
Medical Long Context Reasoning (MLCR-AA, October 2026 methodology)Health & Medicine26.719.4−7.2
Vals Finance IndexIndexes60.967.0+6.2
GDPxProfessional Work5.410.9+5.5
VIALS-AA (pass@3)Multimodal66.570.8+4.3
Finance Agent v2.1Finance & Accounting62.967.2+4.2
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Legal18.823.0+4.2
Vals Legal IndexIndexes71.975.9+4.1
ARC-AGI-1 Semi-PrivateGeneral QA94.598.5+4.0
LMArena Agent Arena Bash ReliabilityAgents2.4-5.2−3.8
AA-LCR v1.1Search82.779.3−3.3
HLAB-Vals-1.1Legal14.217.5+3.3
Tax Agent BenchFinance & Accounting74.976.4+1.5
Vibe Code Bench v1.1Code87.789.1+1.3
LMArena Agent Arena Tool HallucinationAgents0.30.5+0.1

DeepSeek V4.1 Flash 1017vsGemini 4 Carbonnormalised score marginsData Nov 15, 2026Data: Artificial Analysis, Vals AI, Arena and 12 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…