Gemini 4 family vs frontier peers
Gemini 4 Neon, Carbon, Argon and Hydrogen next to Claude Opus 5.5 and GPT-6.2 Sol, using Google DeepMind's launch comparison table plus the independent evals already in our catalog.
| Google launch comparison | Gemini 4 family | Frontier peers | ||||
|---|---|---|---|---|---|---|
| Gemini 4 Neon | Gemini 4 Carbon | Gemini 4 Argon | Gemini 4 Hydrogen | Claude Opus 5.5 | GPT-6.2 Sol | |
| Google DeepMind's published comparison, read 2026-11-19. All figures are the vendor's claims. | ||||||
| Capability indices | 61.0 (Highest reported score in this row) | – | – | – | 59.0 (Second-highest reported score in this row) | 57.0 |
| 70.9% (Highest reported score in this row) | – | – | – | 70.1% (Second-highest reported score in this row) | 66.1% | |
| CursorBench 4.1 | 65.4% (Highest reported score in this row) | – | – | – | 58.1% (Second-highest reported score in this row) | – |
| Terminal-bench 4.1Cheating adjusted accuracy | 65.2% (Second-highest reported score in this row) | – | – | – | 65.8% (Highest reported score in this row) | 61.5% |
| FrontierCode 2.0Main score | 46.2% (Second-highest reported score in this row) | – | – | – | 47.3% (Highest reported score in this row) | 43.8% |
| SWE-sweepProvider harness | 39.0% (Highest reported score in this row) | – | – | – | 10.1% (Second-highest reported score in this row) | 8.3% |
| SWE-sweepMini-SWE harness | 12.8% (Highest reported score in this row) | – | – | – | 3.6% | 11.6% (Second-highest reported score in this row) |
| AutomationBench | 56.3% (Highest reported score in this row) | – | – | – | 42.5% (Second-highest reported score in this row) | 41.3% |
| Agents' Last Exam v1.1 | 75.8% (Highest reported score in this row) | – | – | – | 71.6% | 73.0% (Second-highest reported score in this row) |
| RiemannBench | 84.0% (Highest reported score in this row) | – | – | – | 80.0% (Second-highest reported score in this row) | 80.0% (Second-highest reported score in this row) |
| Terminal-Bench-Science 0.1 | 70.5% (Highest reported score in this row) | – | – | – | 63.3% | 68.0% (Second-highest reported score in this row) |
| HLE-Diamond | 71.3% (Highest reported score in this row) | – | – | – | 67.9% (Second-highest reported score in this row) | 64.0% |
| MKAClimb WSR | 56.5% (Highest reported score in this row) | – | – | – | 53.0% (Second-highest reported score in this row) | 52.7% |
| Chartography | 73.5% (Second-highest reported score in this row) | – | – | – | 66.3% | 75.1% (Highest reported score in this row) |
| LVBench | 89.0% (Highest reported score in this row) | – | – | – | 83.7% | 87.6% (Second-highest reported score in this row) |
| GraphWalks256k to 1M, BFS (F1) | 95.2% (Highest reported score in this row) | – | – | – | 66.8% | 80.1% (Second-highest reported score in this row) |
| CWE-Bench v2 | 70.7% (Highest reported score in this row) | – | – | – | 67.1% (Second-highest reported score in this row) | 66.0% |
| SEC-bench Pro | 90.5% (Highest reported score in this row) | – | – | – | – | 88.3% (Second-highest reported score in this row) |
| WebExploitBench v1.2 | 51.3% (Highest reported score in this row) | – | – | – | 37.1% (Second-highest reported score in this row) | – |
Sources: Gemini 4 Neon launch comparison table (Google DeepMind) (vendor-reported, as of 2026-11-19); catalog benchmarks from Artificial Analysis, Epoch AI. A dash means the source reports no value.
Benchmark data from Epoch AI (CC-BY 4.0).
Updated 2026-11-19.