BALROG BabyAI (VLM): leaderboard
Vision-language models on BabyAI grid-world navigation using visual grid observations, measuring interpretation of rendered game frames for goal completion.
Metric: Progress (%). Source: balrogai.com. Saturation forecast: Estimated already saturated. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet (20241022) | 82 | #291 |
| 2 | Gemini 2.5 Pro (Preview 03-25) | 74 | #123 |
| 3 | Llama 3.2 90B Instruct | 66 | #601 |
| 4 | GPT-4o (2024-05-13) | 62 | #325 |
| 5 | Gemini 1.5 Pro (002) | 58.4 | #349 |
| 6 | Gemini 1.5 Flash (002) | 43.2 | #583 |
| 7 | GPT-4o Mini (2024-07-18) | 38 | #558 |
| 8 | Llama 3.2 11B Instruct | 18 | #1126 |
| 9 | Qwen 2 VL 7B Instruct | 2 | #826 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=balrog-babyai-vlm · How It Works · Data refreshed daily, snapshot 2026-10-20.