- AA Index v5.0.1 61
- Vals Index v2.2 70.9
- CursorBench 4.1 (Score) 65.4
- Terminal-bench 4.1 (Cheating adjusted accuracy) 65.2
- FrontierCode 2.0 (Main score), at the default high thinking level 46.2
- SWE-sweep (Mini-SWE harness). 39.0% with the provider harness, Google Antigravity 12.8
- AutomationBench 56.3
- Agents' Last Exam v1.1 75.8
- RiemannBench 84
- Terminal-Bench-Science 0.1 70.5
- HLE-Diamond 71.3
- MKA (Climb WSR) 56.5
- Chartography 73.5
- LVBench 89
- GraphWalks (256k to 1M, BFS (F1)) 95.2
- CWE-Bench v2 70.7
- SEC-bench Pro 90.5
- WebExploitBench v1.2 (51.26% in the text, 51.3% in the table) 51.3
- Artificial Analysis Coding Agent Index, #1 overall (58.6), text 58.6
- Gray Swan IPI, attack success rate at k=15 attempts 0.0% (lower is better), chart 0
Gemini 4 Neon
Gemini 4 Neon ranks first on 25 of the 80 benchmarks with five or more models scored.
Gemini 4 Neon
Closed weights
- AutomationBench 56.3% #1 /45
- FrontierCode 2.0 46.2% #3 /24
- BioMysteryBench 80% #4 /33
Where has it been number one?
Gemini 4 Neon is number one on 13 benchmarks today.
Source: Official leaderboards
How does it do against its rivals?
Gemini 4 Neon beats Claude Opus 5.5 on 54 of the 68 benchmarks they share. Its closest contest is with Claude Fable 5.2, which is ahead on 26 of 55.
- Claude Opus 5.5 54 ahead, 11 behind, 3 tied Compare
- GPT-6.2 Sol 46 ahead, 10 behind Compare
- GPT-6.1 Sol 54 ahead, 12 behind, 3 tied Compare
- GPT-6 Astra 54 ahead, 16 behind, 3 tied Compare
- GPT-6.2 Astra 42 ahead, 16 behind, 3 tied Compare
- Claude Fable 5.2 28 ahead, 26 behind, 1 tied Compare
Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards
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) | 1,513 | #3 of 345 | Official leaderboard | |
| Extended NYT Connections | 99.1% | #2 of 109 | Official leaderboard | |
| Miðeind Icelandic Popular Culture | 26% | #3 of 61 | Official leaderboard | |
| BullshitBench V2 | 84% | #9 of 137 | Official leaderboard | |
| Miðeind Icelandic Grammar Correction | 80.7% | #7 of 61 | Official leaderboard | |
| Icelandic WikiQA | 68.1% | #7 of 59 | Official leaderboard | |
| Miðeind Icelandic Idioms | 76.8% | #8 of 61 | Official leaderboard | |
| Miðeind Icelandic Terminology | 59.5% | #8 of 61 | Official leaderboard | |
| Miðeind Icelandic Case Government | 98.4% | #9 of 61 | Official leaderboard | |
| Miðeind Icelandic Generative Quality | 78.3% | #11 of 59 | Official leaderboard | |
| GraphWalks | 95.2% | #5 of 29 lab numbers | Lab model card | |
| HLE-Diamond | 71.3% | Not ranked | Lab model card |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SimpleBench | 94.4% | #2 of 110 | Official leaderboard | |
| RandomBench | 0.37 | #3 of 37 | Official leaderboard | |
| BioMysteryBench | 80% | #4 of 33 | Official leaderboard | |
| MysteryMechanism | 45% | #9 of 34 | Official leaderboard | |
| AGI-Pilled Bench | 14.0 | #15 of 19 | Official leaderboard | |
| Riemann-bench | 84% | Not ranked | Lab model card | |
| Terminal-Bench-Science | 70.5% | Not ranked | Lab model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Vibe Code Bench | 93.7% | #2 of 124 | Official leaderboard | |
| Kaggle Pelican Arena | 1,816 | #1 of 26 | Official leaderboard | |
| Brokk Power Ranking | 35.4 | #1 of 24 | Official leaderboard | |
| CursorBench 4.0 | 65.4% | #1 of 12 | Official leaderboard | |
| CursorBench 4.1 | 65.4% | #1 of 12 | Official leaderboard | |
| FrontierCode 2.0 | 46.2% | #3 of 24 | Official leaderboard | |
| AnvilBench | 0.53 | #2 of 12 | Official leaderboard | |
| realitybench | 0.94 | #1 of 5 | Official leaderboard | |
| Ramen Bench | 5 | #5 of 24 | Official leaderboard | |
| Vibe Code Bench 1-100 v1.1 | 42.3% | #3 of 11 | Official leaderboard | |
| AgenticRL-Bench | 10.5% | #5 of 6 | Official leaderboard | |
| Artificial Analysis Coding Agent Index v2 | 58.6 | #1 of 2 lab numbers | Lab model card | |
| SWE-sweep | 12.8% | Not ranked | Lab model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Dubesor AI Chess Leaderboard | 1,907 | #1 of 250 | Official leaderboard | |
| Dubesor Lichess Puzzle Rating | 2,455 | #1 of 62 | Official leaderboard | |
| AutomationBench | 56.3% | #1 of 45 | Official leaderboard | |
| Toloka Arena Enterprise Tool Use (November 2026) | 81% | #1 of 43 | Official leaderboard | |
| Crest Arcade Micro Mouse | 89.3 | #1 of 17 | Official leaderboard | |
| Crest Arcade Nonograms | 91.7% | #1 of 14 | Official leaderboard | |
| Crest Arcade Ravelin | 106.8 | #1 of 14 | Official leaderboard | |
| Crest Arcade Starjar | 114.5 | #1 of 14 | Official leaderboard | |
| Crest Arcade Mines | 100% | #2 of 17 | Official leaderboard | |
| Maze Bench | 17% | #7 of 58 | Official leaderboard | |
| Crest Arcade Wildhaven | 91% | #2 of 14 | Official leaderboard | |
| Crest Arcade Hustle Sim | 70.1 | #3 of 17 | Official leaderboard | |
| Crest Arcade Sudoku | 100% | #3 of 17 | Official leaderboard | |
| Crest Arcade Chess | 1,497 | #4 of 19 | Official leaderboard | |
| Crest Arcade FreeCell | 40% | #4 of 17 | Official leaderboard | |
| RedstoneBench | 63.4 | #4 of 10 | Official leaderboard | |
| Crest Arcade Nations Royale | 87.5 | #7 of 16 | Official leaderboard | |
| Crest Arcade Elemental Armory | 50% | #8 of 17 | Official leaderboard | |
| PokéBench: Tool Assisted Benchmark | 10.0 | #20 of 34 | Official leaderboard | |
| Terminal-Bench 4.1 | 65.2% | Not ranked | Lab model card |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Legal Research Bench | 56.3% | #1 of 97 | Official leaderboard | |
| Vending-Bench 2 | $16,843 | #1 of 70 | Official leaderboard | |
| FoodTruck Bench | $215,319 | #1 of 65 | Official leaderboard | |
| Tuesday Work Index | 71.6 | #1 of 56 | Official leaderboard | |
| Vals Finance Index | 68.3% | #1 of 44 | Official leaderboard | |
| Vals Legal Index | 78.4% | #1 of 42 | Official leaderboard | |
| APEX-Agents 1.1 | 88.9% | #2 of 69 | Official leaderboard | |
| GDP.xlsx | 43.1% | #1 of 34 | Official leaderboard | |
| APEX-Accounting | 70.8% | #2 of 52 | Official leaderboard | |
| DAYJOB: Accounting | 49.6% | #1 of 26 | Official leaderboard | |
| Vals Finance Agent v2.1 | 68.6% | #2 of 44 | Official leaderboard | |
| Vals Index v2.2 | 70.9% | #2 of 41 | Official leaderboard | |
| DualEntry Accounting AI Benchmark | 91.1% | #2 of 29 | Official leaderboard | |
| Argo-Bench | 64.6 | #2 of 24 | Official leaderboard | |
| Excel Modeling Benchmark v1.1 | 76.5% | #4 of 46 | Official leaderboard | |
| DataAnalyticsBench | 43.0 | #5 of 51 | Official leaderboard | |
| Periapical Lesions Bench | 18.7% | #1 of 8 | Official leaderboard | |
| ContractorOS Quantity Takeoff | 51.5% | #6 of 43 | Official leaderboard | |
| Construction's Last Exam | 28.3% | #5 of 30 | Official leaderboard | |
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 19.6% | #8 of 42 | Official leaderboard | |
| DAYJOB: Law | 26.2% | #7 of 22 | Official leaderboard | |
| EEBench | 63% | #13 of 34 | Official leaderboard | |
| Wojak-Bench 3D | 1,054 | #4 of 10 | Official leaderboard | |
| MKA | 56.5% | #1 of 2 lab numbers | Lab model card |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Humanity's Sixth Sense | 75.9% | #1 of 38 | Official leaderboard | |
| BuseyBench SVG Test | 7.10 | #9 of 261 | Official leaderboard | |
| MineBench | 2,085 | #6 of 81 | Official leaderboard | |
| Chartography | 86.6% | #6 of 56 | Official leaderboard | |
| HearingBench | 45% | #2 of 16 | Official leaderboard | |
| LVBench | 89% | Not ranked | Lab model card |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CWE-bench 2 | 70.7% | #2 of 20 | Official leaderboard | |
| AutoBaxBench | 71.4% | #3 of 18 | Official leaderboard | |
| Gray Swan IPI Arena benchmark | 0% | Not ranked | Lab model card | |
| SEC-bench Pro | 90.5% | Not ranked | Lab model card | |
| WebExploitBench | 51.3% | Not ranked | Lab model card |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Parallel Search Capability Leaderboard | 88.1 | #1 of 51 | Official leaderboard | |
| LiveBench | 87.4% | #1 of 36 | Official leaderboard | |
| Artificial Analysis Intelligence Index v5 | 61.4 | #3 of 69 | Official leaderboard | |
| Vals Science Index | 62.6% | #4 of 28 | Official leaderboard |
What does it cost to run?
On the 54 leaderboards that publish a price, its best run usually costs 2.8 times the median model's. On 23 of them, no model comes close to its score for less.
| Benchmark | Score | Price | From cheapest to priciest | Against the median | Close to its score for less Close for less |
|---|---|---|---|---|---|
| BuseyBench SVG Test | 7.10 | $0.7 56× the median | 56× | GPT-6.1 Terra 7.30 for $0.19 | |
| Dubesor AI Chess Leaderboard | 1,907 | $2.22 11× the median | 11× | | |
| BullshitBench V2 | 84% | $0.056 12× the median | 12× | | |
| Vibe Code Bench | 93.7% | $6.76 1.7× the median | 1.7× | None | |
| Legal Research Bench | 56.3% | $7.37 6.7× the median | 6.7× | | |
| SimpleBench | 94.4% | $7.99 per run 2.7× the median | 2.7× | Claude Fable 5.2 94.9% for $5.15 | |
| Vending-Bench 2 | $16,843 | $274 per run 2.1× the median | 2.1× | None | |
| FoodTruck Bench | $215,319 | $30.75 per run 17× the median | 17× | None | |
| Dubesor Lichess Puzzle Rating | 2,455 | $0.16 13× the median | 13× | None | |
| Maze Bench | 17% | $206 per run 4.0× the median | 4.0× | GPT-6.1 Sol 15% for $63.24 | |
| DataAnalyticsBench | 43.0 | $17.03 per run 5.4× the median | 5.4× | GPT-6.1 Sol 43.0 for $2.61 | |
| Parallel Search Capability Leaderboard | 88.1 | $206 1.6× the median | 1.6× | None | |
| MineBench | 2,085 | $0.26 0.4× the median | 0.4× | None | |
| AutomationBench | 56.3% | $1.69 3.1× the median | 3.1× | None | |
| Excel Modeling Benchmark v1.1 | 76.5% | $3.82 2.4× the median | 2.4× | GPT-6.2 Astra 76% for $2.80 | |
| Toloka Arena Enterprise Tool Use (November 2026) | 81% | $4.54 2.1× the median | 2.1× | None | |
| Vals Finance Agent v2.1 | 68.6% | $5.85 3.6× the median | 3.6× | | |
| Vals Finance Index | 68.3% | $4.88 3.5× the median | 3.5× | None | |
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 19.6% | $5.97 1.8× the median | 1.8× | | |
| Vals Legal Index | 78.4% | $6.10 2.6× the median | 2.6× | | |
| Vals Index v2.2 | 70.9% | $7.40 2.0× the median | 2.0× | None | |
| LiveBench | 87.4% | $2.32 2.6× the median | 2.6× | None | |
| EEBench | 63% | $2.58 1.7× the median | 1.7× | GPT-6.1 Sol 63.6% for $0.3 | |
| MysteryMechanism | 45% | $6.63 5.1× the median | 5.1× | GPT-6.1 Sol 46.4% for $0.28 | |
| BioMysteryBench | 80% | $4.41 4.3× the median | 4.3× | GPT-6.1 Sol 79.6% for $0.37 | |
| Construction's Last Exam | 28.3% | $4.20 per run 2.4× the median | 2.4× | GPT-6.1 Terra 36.4% for $1.40 | |
| PokéBench: Tool Assisted Benchmark | 10.0 | $7.06 per run 0.9× the median | 0.9× | | |
| Vals Science Index | 62.6% | $8.26 2.5× the median | 2.5× | GPT-6.2 Astra 68.4% for $4.76 | |
| DAYJOB: Accounting | 49.6% | $3.48 1.0× the median | 1.0× | None | |
| DualEntry Accounting AI Benchmark | 91.1% | $0.17 1.9× the median | 1.9× | | |
| Argo-Bench | 64.6 | $5.79 2.3× the median | 2.3× | None | |
| Brokk Power Ranking | 35.4 | $21.88 1.8× the median | 1.8× | None | |
| FrontierCode 2.0 | 46.2% | $4.38 0.9× the median | 0.9× | None | |
| Ramen Bench | 5 | $3.24 3.5× the median | 3.5× | GPT-6 Astra 5 for $1.86 | |
| DAYJOB: Law | 26.2% | $2.35 0.8× the median | 0.8× | GPT-6.2 Sol 32.7% for $1.42 | |
| CWE-bench 2 | 70.7% | $6.52 3.3× the median | 3.3× | None | |
| Crest Arcade Chess | 1,497 | $2.19 2.5× the median | 2.5× | | |
| Crest Arcade Elemental Armory | 50% | $0.15 3.3× the median | 3.3× | GPT-6 Luna 51.3% for $0.0016 | |
| Crest Arcade FreeCell | 40% | $0.53 1.4× the median | 1.4× | Claude Haiku 5.5 63.3% for $0.071 | |
| Crest Arcade Hustle Sim | 70.1 | $3.70 11× the median | 11× | | |
| Crest Arcade Micro Mouse | 89.3 | $1.58 2.9× the median | 2.9× | None | |
| Crest Arcade Mines | 100% | $2.06 4.4× the median | 4.4× | | |
| Crest Arcade Sudoku | 100% | $0.3 1.8× the median | 1.8× | GPT-6.2 Astra 100% for $0.2 | |
| CursorBench 4.0 | 65.4% | $10.67 1.2× the median | 1.2× | None | |
| CursorBench 4.1 | 65.4% | $10.67 1.2× the median | 1.2× | None | |
| Crest Arcade Nations Royale | 87.5 | $2.91 3.3× the median | 3.3× | GPT-6.1 Terra 91.7 for $0.29 | |
| Vibe Code Bench 1-100 v1.1 | 42.3% | $44.11 3.3× the median | 3.3× | GPT-6.2 Sol 40.9% for $12.74 | |
| Crest Arcade Nonograms | 91.7% | $0.61 3.4× the median | 3.4× | None | |
| Crest Arcade Ravelin | 106.8 | $0.89 6.4× the median | 6.4× | None | |
| Crest Arcade Starjar | 114.5 | $6.45 6.8× the median | 6.8× | GPT-6 Astra 112.3 for $3.56 | |
| Crest Arcade Wildhaven | 91% | $0.41 10.0× the median | 10.0× | None | |
| Extended NYT Connections | 99.1% | $0.071 2.5× the median | 2.5× | GPT-6.2 Astra 98.2% for $0.015 | |
| Periapical Lesions Bench | 18.7% | $1.97 4.6× the median | 4.6× | None | |
| AGI-Pilled Bench | 14.0 | $1.08 19× the median | 19× | GLM-5.3 8.00 for $0.058 |
Inside agents
Boards that list Gemini 4 Neon inside an agent or product, such as OpenCode + Gemini 4 Neon. Those scores belong to the agent, so we don't count them as the model's own. Its own score on the same board is beside each one.
| Benchmark | Entry | Agent's score | Its own score |
|---|---|---|---|
| InferenceBench | OpenCode + Gemini 4 Neon | 11.0× |
Its lab's numbers put it at 174.8, number 3 of 412
The Self-Reported Index fits one score to the 5 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 170.1 to 179.5. How we compute it.
- GraphWalks 95.2
- LVBench 89
- Chartography 73.5
- Terminal-Bench-Science 70.5
- AutomationBench 56.3
Who measured it?
Independent runs cover 79 of its 91 benchmarks. For the other 12, we have only Google DeepMind's own number or a claim Epoch AI collected.
- An independent run and a claim 7
- An independent run only 72
- A claim only 12
Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards
Its lab's numbers next to 7 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 |
|---|---|---|---|
| Chartography | 73.5% Chartography | 86.6% Leaderboard, high | −13 pts |
| Artificial Analysis Intelligence Index v5 | 61.0 AA Index v5.0.1 | 61.4 Leaderboard, xhigh | −0.4 |
| FrontierCode 2.0 | 46.2% FrontierCode 2.0 (Main score), at the default high thinking level | 46.2% Leaderboard, high, antigravity-cli | Same |
| Vals Index v2.2 | 70.9% Vals Index v2.2 | 70.9% Leaderboard | Same |
| AutomationBench | 56.3% AutomationBench | 56.3% Leaderboard, xhigh | Same |
| CursorBench 4.1 | 65.4% CursorBench 4.1 (Score) | 65.4% Leaderboard, xhigh, Cursor | Same |
| CWE-bench 2 | 70.7% CWE-Bench v2 | 70.7% Leaderboard, high reasoning; 1h limit, Antigravity | Same |
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.
Its system card
Every result in the longer documents the lab published with it.
Capabilities
- Artificial Analysis Intelligence Index v5 61
- Vals Index v2.2 70.9
- CursorBench 4.1 65.4
- Terminal-Bench 4.1 65.2
- FrontierCode 2.0 46.2
- SWE-sweep 12.8
- AutomationBench 56.3
- Agents' Last Exam v1.1 75.8
- Riemann-bench 84
- Terminal-Bench-Science 70.5
- HLE-Diamond 71.3
- MKA 56.5
- Chartography 73.5
- LVBench 89
- GraphWalks 95.2
Cyber risk
- CWE-bench 2 70.7
- SEC-bench Pro 90.5
- WebExploitBench 51.3