# SIMA 2: A Generalist Embodied Agent for Virtual Worlds

*Source* · `src-deepmind-2025-sima-2-paper` · status: **n/a**

| field | value |
|---|---|
| kind | Preprint |
| authors | SIMA Team (Google DeepMind) |
| venue | arXiv |
| date | 2025-12-04 |
| url | https://arxiv.org/abs/2512.04797 |
| arxiv | 2512.04797 |
| accessed | 2026-10-24 |

## Notes

Roughly doubles SIMA 1 average success, approaching humans; over 10% better than SIMA 1 on held-out ASKA and MineDojo; 26 of 50 MineDojo categories. Task setter plus reward model scoring 0-100 (>=50 success). Gemini Flash-Lite base for latency; SFT cut AIME about 25.5%; weak combat; short memory.

## Backlinks (computed)

**cited by:** [Generalist instructable game agents (SIMA)](../technique/generalist-game-agents.md)

## Citation

SIMA Team (Google DeepMind) (2025-12-04). *SIMA 2: A Generalist Embodied Agent for Virtual Worlds*. arXiv https://arxiv.org/abs/2512.04797

---
Provenance: created in run `2026-10-24-techniques-llm-agents`.