VideoGameBench: leaderboard
Tests VG-Agent models on ten test games from a 23-game dev/test suite using raw visual inputs. Mean checkpoint progress in percent; games continue during inference. The paused three-game Lite protocol is a separate board.
Metric: Mean game progress (%; VG-Agent; real-time; 10 test games). Source: vgbench.com. Saturation forecast: Around 2031. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro (Preview 03-25) | 0.48 | #123 |
| 2 | GPT-4o (2024-08-06) | 0.09 | #336 |
| 3 | Gemini 2.0 Flash | 0 | #342 |
| 4 | Llama 4 Maverick Instruct FP8 | 0 | #414 |
| 5 | Claude 3.7 Sonnet (20250219) | 0 | #201 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=videogamebench · How It Works · Data refreshed daily, snapshot 2026-10-20.