- AA Index v5.0 59
- Vals Index v2.2 70.4
- SWE-sweep 9.6
- FrontierSWE 2 61.2
- Terminal-Bench 4.0 (AA) 65.2
- Agents' Last Exam v1.1 (Score) 77.4
- OSWorld-2.1 (Offline subset) 81.5
- AutomationBench (Zapier) 53.9
- Vending-Bench 2 (Bank balance $) 16484
- Chartography 72.5
- Blueprint-Bench 2 (Similarity score) 58.6
- GraphWalks (256k to 1M, BFS (F1)) 96.2
- RiemannBench 80
- Terminal-Bench-Science 0.1 (AA) 64.8
- CWE-bench 71
- ExploitGym 54.8
- SEC-bench Pro 88.2
- WebDev Arena (#4, 1768, text) 1768
- Gray Swan Indirect Prompt Injection (IPI) Benchmark, attack success rate 0.0% (lower is better) 0
Gemini 4 Carbon
Gemini 4 Carbon ranks first on 40 of the 111 benchmarks with five or more models scored.
115
Benchmarks with a score
40
Ranked first, out of five or more models
67
In the top three
Google DeepMind Oct 2026 Gemini 4 Carbon
Closed weights
169.8
#2 of 256
Epoch Capabilities Index
Self-Reported Index 173.2 #2 of 404
- AutomationBench 53.9% #1 /42
- GPQA Diamond 96.5% #1 /229
- FrontierCode 2.0 42.6% #4 /32
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LMArena (Elo) | 1,526 | #1 of 335 | Official leaderboard | |
| Extended NYT Connections | 99.3% | #1 of 103 | Official leaderboard | |
| SimpleQA Verified | 77.7% | #2 of 94 | Epoch AI run | |
| FACTS Benchmark Suite | 78.3% | #1 of 23 | Official leaderboard | |
| FACTS Multimodal | 60.2% | #1 of 23 | Official leaderboard | |
| FACTS Parametric V2 | 81.6% | #3 of 27 | Official leaderboard | |
| GraphWalks | 96.2% | #4 of 27 | Lab model card | |
| HLE-Diamond | 63.6% | #3 of 20 | Official leaderboard | |
| BullshitBench V2 | 72% | #21 of 129 | Official leaderboard | |
| FACTS Grounding v2 | 77.2% | #6 of 25 | Official leaderboard | |
| Joke Arena | 1,395 | #25 of 25 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GPQA Diamond | 96.5% | #1 of 229 | Epoch AI run | |
| SimpleBench | 92.6% | #2 of 106 | Official leaderboard | |
| ARC-AGI-2 | 94.2% | #3 of 98 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 95.1% | #4 of 68 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 90.5% | #6 of 86 | Official leaderboard | |
| OTIS Mock AIME 2024-2025 | 100% | #15 of 212 | Epoch AI run | |
| ARC-AGI | 98.5% | #7 of 95 | Official leaderboard | |
| Terminal-Bench-Science | 64.8% | #2 of 12 | Lab model card | |
| Riemann-bench | 80% | #4 of 20 | Official leaderboard | |
| FrontierMath Erdős | 2.94% | #4 of 15 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Code Arena | 1,768 | #4 of 128 | Official leaderboard | |
| ALE-Bench | 2,929 | #4 of 126 | Official leaderboard | |
| CursorBench | 60.4% | #1 of 21 | cursor.com, via Epoch AI | |
| CursorBench 4.0 | 60.4% | #1 of 21 | Official leaderboard | |
| SWE-Marathon v1.1 | 53.8% | #1 of 20 | Official leaderboard | |
| DesignArena 3D Design | 1,419 | #8 of 136 | Official leaderboard | |
| DesignArena UI Component | 1,349 | #10 of 139 | Official leaderboard | |
| Vibe Code Bench | 89.1% | #9 of 120 | Official leaderboard | |
| DesignArena Agentic Overall | 1,307 | #6 of 74 | Official leaderboard | |
| Floorplan Arena | 1,792 | #1 of 11 | Official leaderboard | |
| DesignArena Code | 1,331 | #14 of 143 | Official leaderboard | |
| FrontierCode 2.0 | 42.6% | #4 of 32 | Official leaderboard | |
| DesignArena Game Dev | 1,322 | #19 of 143 | Official leaderboard | |
| SWE Atlas v1.1 | 71.4% | #4 of 28 | Official leaderboard | |
| DesignArena Website | 1,293 | #23 of 151 | Official leaderboard | |
| FrontierCS 2 | 54.2% | #3 of 19 | Official leaderboard | |
| FrontierSWE | 61.2% | #5 of 25 | Official leaderboard | |
| Vals RSI Index | 0.28 | #7 of 27 | Official leaderboard | |
| PostTrainBench 1.2 | 41% | #6 of 21 | Official leaderboard | |
| Android Bench 2.0 | 23.3% | #7 of 17 | Official leaderboard | |
| OWE-Bench | 32.8% | #6 of 14 | Official leaderboard | |
| DesignArena Image to HTML | 1,211 | #23 of 51 | Official leaderboard | |
| WebDev Arena | 1,768 | #1 of 2 | Lab model card | |
| Chart Arena | 1,506 | #8 of 15 | Official leaderboard | |
| Landing Page Arena | 1,465 | #9 of 14 | Official leaderboard | |
| SWE-sweep | 9.60% | #1 of 1 | Lab model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Chess Puzzles | 78% | #2 of 93 | Epoch AI run | |
| Game Arena Four in a Row | 1,003 | #1 of 43 | Official leaderboard | |
| AutomationBench | 53.9% | #1 of 42 | Official leaderboard | |
| Game Arena Chess | 1,646 | #1 of 42 | Official leaderboard | |
| Kaggle Game Arena | 638 | #1 of 42 | Official leaderboard | |
| Game Arena Checkers | 1,382 | #1 of 41 | Official leaderboard | |
| Game Arena Word Association | 734 | #1 of 40 | Official leaderboard | |
| Game Arena Bargaining | 317 | #1 of 39 | Official leaderboard | |
| Game Arena Clobber | 819 | #1 of 39 | Official leaderboard | |
| Game Arena Lines of Action | 1,274 | #1 of 39 | Official leaderboard | |
| Game Arena Nine Men's Morris | 1,190 | #1 of 39 | Official leaderboard | |
| Game Arena Reversi | 1,306 | #1 of 39 | Official leaderboard | |
| Game Arena Ultimate Tic-Tac-Toe | 720 | #1 of 39 | Official leaderboard | |
| Game Arena Word Art | 2,528 | #1 of 39 | Official leaderboard | |
| Game Arena Dark Hex | 552 | #2 of 41 | Official leaderboard | |
| Game Arena Coin Game | 506 | #2 of 39 | Official leaderboard | |
| Game Arena Go | 1,830 | #2 of 39 | Official leaderboard | |
| Agent Arena | 17% | #3 of 56 | Official leaderboard | |
| Mystery Game Puzzles | 73% | #5 of 84 | Epoch AI run | |
| EBR-bench | 87.6% | #2 of 31 | Epoch AI run | |
| ARC-AGI-3 | 99.1% | #2 of 27 | Official leaderboard | |
| Terminal-Bench 4.0 | 65.2% | #5 of 28 | Lab model card | |
| OSWorld 2.1 (offline subset) | 81.5% | #2 of 3 | Lab model card | |
| Game Arena Dots and Boxes | 487 | #39 of 40 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Legal Research Bench | 55.8% | #1 of 93 | Official leaderboard | |
| Vending-Bench 2 | $16,484 | #1 of 67 | Official leaderboard | |
| APEX-Agents 1.1 | 90.4% | #1 of 65 | Official leaderboard | |
| APEX-Agents | 84.2% | #1 of 61 | Collected by Epoch AI, source not named | |
| FoodTruck Bench | $77,218 | #1 of 59 | Official leaderboard | |
| Tax Agent Bench | 51.8% | #2 of 84 | Official leaderboard | |
| Vals Finance Index | 67% | #1 of 40 | Official leaderboard | |
| DesignArena PDF | 1,335 | #1 of 37 | Official leaderboard | |
| APEX-1 | 80% | #2 of 71 | Official leaderboard | |
| Vending-Bench Arena | $15,962 | #1 of 32 | Official leaderboard | |
| EEBench | 83.9% | #1 of 30 | Official leaderboard | |
| YC-Bench | $3,425,290 | #2 of 59 | Official leaderboard | |
| APEX-Accounting | 70.7% | #2 of 49 | Official leaderboard | |
| Excel Modeling Benchmark v1.1 | 76.5% | #3 of 42 | Official leaderboard | |
| Agents' Last Exam v1.1 | 45.9% | #1 of 14 | Official leaderboard | |
| Vals Legal Index | 75.9% | #3 of 38 | Official leaderboard | |
| Vals Index v2.2 | 70.5% | #3 of 37 | Official leaderboard | |
| DeepFinance | 56.6% | #2 of 23 | Official leaderboard | |
| PRBench | 53.7% | #5 of 48 | Official leaderboard | |
| GDPx | 10.9% | #3 of 20 | Official leaderboard | |
| Vals Finance Agent v2.1 | 67.2% | #7 of 40 | Official leaderboard | |
| TaxCalcBench v2 | 71.4% | #7 of 32 | Official leaderboard | |
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 17.5% | #9 of 38 | Official leaderboard | |
| RWE-bench | 20% | #5 of 15 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Blueprint-Bench 2 | 58.6% | #1 of 38 | Official leaderboard | |
| MineBench | 2,195 | #2 of 75 | Official leaderboard | |
| Humanity's Sixth Sense | 71.1% | #1 of 34 | Official leaderboard | |
| Code2Video Arena | 1,599 | #1 of 13 | Official leaderboard | |
| Furniture Assembly Benchmark | 77.5% | #6 of 35 | Official leaderboard | |
| OmniParseBench | 95.6 | #2 of 9 | Official leaderboard | |
| Chartography | 72.5% | #7 of 14 | Lab model card |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| TxBench-Preclinical-Pharmacology | 66% | #1 of 30 | Official leaderboard | |
| TxBench-Antibody-Discovery | 65% | #2 of 30 | Official leaderboard | |
| TxBench-Oligo-Discovery | 55.8% | #2 of 30 | Official leaderboard | |
| ExploitGym | 54.8% | #1 of 15 | Lab model card | |
| CWE-bench | 71% | #1 of 12 | Official leaderboard | |
| Gray Swan IPI Arena benchmark | 0% | #1 of 11 | Lab model card | |
| SEC-bench Pro | 88.2% | #1 of 11 | Lab model card | |
| DeepsecBench | 35.2 | #7 of 29 | Official leaderboard | |
| FORTRESS | 16.4% | #20 of 73 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CAIS Text Capabilities Index | 59.0 | #5 of 68 | Official leaderboard | |
| CAIS Risk Index | 33.4 | #8 of 67 | Official leaderboard | |
| LiveBench | 83.5% | #4 of 31 | Official leaderboard | |
| Artificial Analysis Intelligence Index v5 | 59.0 | #1 of 2 | Lab model card |
Its lab's numbers put it at 173.2, 3.4 points above its ECI
The Self-Reported Index fits one score to the 6 capability results Google DeepMind reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 169.5 to 176.9. How we compute it.
- GraphWalks 96.2
- Chartography 72.5
- Terminal-Bench 4.0 65.2
- Terminal-Bench-Science 64.8
- FrontierSWE 61.2
- AutomationBench 53.9
Its lab's numbers next to 8 leaderboards
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
| Benchmark | Lab says | Board says | Gap |
|---|---|---|---|
| Agents' Last Exam v1.1 | 77.4% Agents' Last Exam v1.1 (Score) | 45.9% Leaderboard, high, ALE-Claw | +32 pts |
| Vals Index v2.2 | 70.4% Vals Index v2.2 | 70.5% Leaderboard | Same |
| AutomationBench | 53.9% AutomationBench (Zapier) | 53.9% Leaderboard, high | Same |
| Vending-Bench 2 | $16,484 Vending-Bench 2 (Bank balance $) | $16,484 Leaderboard | Same |
| Blueprint-Bench 2 | 58.6% Blueprint-Bench 2 (Similarity score) | 58.6% Leaderboard | Same |
| CWE-bench | 71% CWE-bench | 71% Leaderboard, Antigravity | Same |
| FrontierSWE | 61.2% FrontierSWE 2 | 61.2% Leaderboard, mean@5, proximus | Same |
| Riemann-bench | 80% RiemannBench | 80% Leaderboard, high | Same |
Google DeepMind's models over time
Google DeepMind's models in color, against every other model on the Epoch Capabilities Index, by release date.
Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.