# Playable Neural World Models

Labs increasingly ship generative 'world models' that simulate interactive environments in real time from learned dynamics rather than hand-coded engines, moving neural simulation from video generation toward playable, controllable environments usable for games, training data, and embodied-agent simulation; expect recurring releases pushing fidelity, frame rate, and openness.

- Conviction: 33 / 100 (forming)
- Horizon: Emerging (watchlist)
- Tracking since: 2026-07-09T00:00:00.000Z
- Last updated: 2026-07-09T05:39:12.486Z
- Canonical: https://polylog.news/ai/trends/interactive-world-models
- Publisher: Polylog
- Affected regions: United States

## Recent evidence

- [confirms] General Intuition, Kyutai and Epic Games Open-Source a Playable Rocket League World Model (2026-07-09): General Intuition, Kyutai and Epic Games open-sourced MIRA, a 5B-parameter latent-diffusion world model that simulates full 2v2 Rocket League matches at 20fps on a single GPU, trained on 10,000 hours of bot self-play.