Meta released a stateful first-person 3D simulator and InteractiveBench with 120 frozen multi-turn tasks across six categories. Success is checked against world state, but the rubric still requires human adjudication. Reproduction needs Linux, an NVIDIA GPU, about 100 GB of assets and separately licensed Isaac Sim, BEHAVIOR, MANO and AMASS components.
Meta opens InteractiveWorld and a state-verifiable multimodal benchmark
2026-10-24Frontier watchWATCH
Meta released a stateful first-person 3D simulator and InteractiveBench with 120 frozen multi-turn tasks across six categories. Success is checked against world state, but the rubric still requires human adjudication. Reproduction needs Linux, an NVIDIA GPU, about 100 GB of assets and separately licensed Isaac Sim, BEHAVIOR, MANO and AMASS components.
What this means for you
State-based checks can expose whether a multimodal agent actually changed the environment rather than narrated success. The non-commercial license, gated assets and high setup cost make it a watch item until cheaper reproduction or transfer evidence appears.
FRONTIER WATCH
Public artifacts and validated results are different.
The recorded fact is the public work and its artifacts. Author-reported outcomes do not establish independent reproduction. Next checks include usable code, task conditions and third-party validation.
Evidence & related material
Technical details & scores
| Benchmark | InteractiveBench, 120 frozen multi-turn tasks in six categories |
|---|---|
| Verification | World-state checks plus required human adjudication |
| License | Repository CC BY-NC 4.0; dependencies and gated assets have separate terms |
| Requirements | Linux, NVIDIA GPU, about 100 GB assets; tested on A100 80 GB |
| Time Basis | Official GitHub repository created at 2026-10-24T02:51:09Z; recovered outside overlap as late discovery |
Technical impact 3
Engineering value 3
Adoption signal 2
Maturity 2
Verification cost 5
Risk 4