Enter opens the highlighted result. Esc closes.

Google Gemini 4 Carbon at launch

  • Google
  • Released November 10, 2026
  • 1.05M context

Google released Gemini 4 Carbon on November 10, 2026. As of November 14, 2026, it ranks #2 of 118 models on the LLMBenchmark Index, between Anthropic Claude Fable 5.2 (#1) and Anthropic Claude Opus 5.5 (#3). Google's own announcement reports 20 benchmark results, including AutomationBench (53.9%) and Chartography (72.5). It costs $2.00 per million input tokens and $10.00 per million output tokens on OpenRouter.

Benchmarks and pricing API providers

What Google reported at launch

From “Gemini 4 Carbon: our next frontier model”, October 28, 2026. The maker's own numbers, not re-run by us.

Show all 20

Independent results so far

73 scores from 73 boards that test models themselves. Ranks compare Gemini 4 Carbon with every model on the same board.

General (16)

BenchmarkScoreRankSettingSource
LiveBench data analysis 86.5% #1 of 31 — LiveBench
LMArena Text 1538 #1 of 151 High LMArena (Hugging Face dataset)
LMArena Text: English 1530 #1 of 151 High LMArena (Hugging Face dataset)
LMArena Text: entertainment and media 1527 #1 of 151 High LMArena (Hugging Face dataset)
LMArena Text: instruction following 1539 #1 of 151 High LMArena (Hugging Face dataset)
LMArena Text: longer queries 1549 #1 of 151 High LMArena (Hugging Face dataset)
LMArena Text: multi-turn 1569 #1 of 151 High LMArena (Hugging Face dataset)
LMArena Text: non-English 1534 #1 of 151 High LMArena (Hugging Face dataset)
Show all 16
LMArena Text: Russian 1560 #1 of 151 High LMArena (Hugging Face dataset)
LMArena Text: writing and language 1542 #1 of 151 High LMArena (Hugging Face dataset)
LMArena Text: Chinese 1613 #2 of 146 High LMArena (Hugging Face dataset)
LMArena Text: creative writing 1533 #2 of 151 High LMArena (Hugging Face dataset)
Vals Index 70.5% #3 of 34 — Vals Index
Furniture assembly 77.5% #6 of 36 High Epoch AI benchmarking hub
GDPval-AA 1733 #8 of 117 High Artificial Analysis
Long context reasoning (AA-LCR) 79.3% #54 of 176 High Artificial Analysis

Coding (13)

BenchmarkScoreRankSettingSource
CursorBench 4.0 60.4% #1 of 20 high CursorBench (Cursor)
LiveBench coding 91.0% #1 of 31 — LiveBench
LMArena Text: coding 1564 #1 of 151 High LMArena (Hugging Face dataset)
LMArena Text: software and IT 1552 #1 of 151 High LMArena (Hugging Face dataset)
Terminal-Bench 4.0 (Artificial Analysis run) 65.2% #2 of 111 High Artificial Analysis
LMArena WebDev: HTML 1770 #3 of 104 High LMArena (Hugging Face dataset)
LMArena WebDev 1768 #4 of 105 High LMArena WebDev
LMArena WebDev: React 1792 #4 of 97 High LMArena (Hugging Face dataset)
Show all 13
FrontierSWE 61.2% #5 of 22 High Epoch AI benchmarking hub
Terminal-Bench 4 (Vals AI) 63.1% #5 of 57 Mini-SWE-agent Vals AI boards
ALE-Bench 2374.7 #6 of 110 High ALE-Bench (Sakana AI)
FrontierCode 2.0 40.6% #7 of 55 High Artificial Analysis
Vibe Code Bench (Vals AI) 89.1% #9 of 105 OpenHands Vals AI boards

Reasoning (13)

BenchmarkScoreRankSettingSource
ARC-AGI 98.5% #1 of 75 Medium Epoch AI benchmarking hub
GPQA Diamond (Epoch AI run) 96.5% #1 of 115 High Epoch AI benchmarking hub
LMArena Text: expert 1582 #1 of 150 High LMArena (Hugging Face dataset)
LMArena Text: hard prompts 1551 #1 of 151 High LMArena (Hugging Face dataset)
LMArena Text: hard prompts in English 1545 #1 of 151 High LMArena (Hugging Face dataset)
ARC-AGI-2 94.2% #2 of 76 Medium Epoch AI benchmarking hub
Chess puzzles 78.0% #2 of 81 High Epoch AI benchmarking hub
EBR-bench 87.6% #2 of 29 High Epoch AI benchmarking hub
Show all 13
HLE-Diamond (Artificial Analysis run) 69.1% #2 of 74 High Artificial Analysis
SimpleBench 92.6% #2 of 72 — Epoch AI benchmarking hub
CritPt v2, hard subset (Artificial Analysis run) 69.0% #4 of 56 High Artificial Analysis
Mystery game puzzles 73.0% #5 of 75 High Epoch AI benchmarking hub
LiveBench reasoning 90.6% #10 of 31 — LiveBench

Agentic work (10)

BenchmarkScoreRankSettingSource
APEX-Agents 84.2% #1 of 59 — Epoch AI benchmarking hub
Legal Research Bench (Vals AI) 55.8% #1 of 83 — Vals AI boards
Vending-Bench 2 $16,484 #1 of 62 — Epoch AI benchmarking hub
LiveBench research 92.5% #2 of 31 — LiveBench
Tax Agent Bench (Vals AI) 51.8% #2 of 77 — Vals AI boards
Design Arena: agents building mobile apps 1284 #4 of 55 — Design Arena
LiveBench long agentic 72.7% #5 of 31 — LiveBench
Harvey's Legal Agent Benchmark (Vals AI) 17.5% #7 of 34 — Vals AI boards
Show all 10
LiveBench browser 80.0% #10 of 31 — LiveBench
Design Arena: spreadsheets 1224 #15 of 39 — Design Arena

Knowledge (8)

BenchmarkScoreRankSettingSource
FACTS Benchmark Suite 78.4% #1 of 23 — Kaggle Benchmarks
LMArena Text: business and finance 1520 #1 of 151 High LMArena (Hugging Face dataset)
LMArena Text: legal and government 1542 #1 of 150 High LMArena (Hugging Face dataset)
AA-Omniscience 53.5 #2 of 58 High Artificial Analysis
LMArena Text: science 1535 #2 of 151 High LMArena (Hugging Face dataset)
SimpleQA Verified 77.7% #2 of 77 High Epoch AI benchmarking hub
LMArena Text: medicine and healthcare 1525 #4 of 150 High LMArena (Hugging Face dataset)
PRBench Legal 53.3% #4 of 39 — Scale SEAL leaderboards

Web design (6)

BenchmarkScoreRankSettingSource
Design Arena: 3D 1434 #5 of 121 — Design Arena
LiveBench design 80.2% #5 of 31 — LiveBench
Design Arena: UI components 1357 #7 of 124 — Design Arena
Design Arena: all code categories 1348 #9 of 129 — Design Arena
Design Arena: websites 1319 #9 of 134 — Design Arena
Design Arena: game dev 1326 #16 of 128 — Design Arena

Math (5)

BenchmarkScoreRankSettingSource
MathArena 85.4% #1 of 72 high MathArena
OTIS Mock AIME 100.0% #1 of 115 High Epoch AI benchmarking hub
FrontierMath, tier 4 95.1% #4 of 66 High Epoch AI benchmarking hub
LMArena Text: mathematical work 1538 #4 of 147 High LMArena (Hugging Face dataset)
FrontierMath, tiers 1 to 3 90.5% #6 of 80 High Epoch AI benchmarking hub

Intelligence (2)

BenchmarkScoreRankSettingSource
AA Intelligence Index 60.0 #2 of 55 — Artificial Analysis (via OpenRouter)
Epoch Capabilities Index 169.8 #2 of 125 — Epoch AI benchmarking hub