Supported

What Robo Use 0.1.1 supports.

Models#

Model	Harness
`claude-sonnet-5`	`claude-code`
`gpt-6-astra`	`codex`
`zai-org/GLM-5.3` on Baseten	`claude-code-glm`, `codex-glm`, `mini-swe-agent-glm`
`allenai/MolmoAct2-LIBERO-LeRobot`	`molmoact2`

Harnesses#

Harness
`oracle`, `noop`
`claude-code`
`codex`
`claude-code-glm`, `codex-glm`, `mini-swe-agent-glm`
`molmoact2`

What each needs: Harnesses.

Simulators#

Backend	Suites	Tasks	Install
tabletop (MuJoCo)	`libero-style`, `arc-style`, `robo-use-families`, `safety`, `hard`, part of `vision`	95	`[sim]`
Meta-World 3.1	`metaworld`, part of `vision`	74	`[sim,metaworld]`
Gymnasium-Robotics	`gymrobotics`	12	`[gymrobotics]`
robosuite 1.5	`robosuite`	21	`[robosuite]`
LIBERO	`libero`	20	by hand
MuJoCo Menagerie	`menagerie`	15	`[sim]` + `robouse fetch-assets`
RoboHarm scenes	`roboharm`	80	`[sim]` + `robouse fetch-assets`
Skydio X2 drone	`drone`	10	`[sim]` + `robouse fetch-assets`
DexJoCo	`dexjoco`	14	`[sim]` + DexJoCo, see Suites

Per-suite results: Suites.

Platforms#

Platform	Status
macOS on Apple silicon, Python 3.11, 3.12, 3.13, MuJoCo 3.3.0	supported
Python 3.14	not supported: MuJoCo 3.3.0 and Meta-World have no build for it

Isolation#

BenchFlow is the environment framework whose task and trial formats Robo Use uses (Robo Use and BenchFlow). The released runner does not start BenchFlow; exported tasks run under bench eval run.

Mode	Status
Local runner, macOS: model harnesses under `sandbox-exec`, file access denied to `ROBOUSE_SANDBOX_DENY` paths	supported
Local runner, Linux	no sandbox
BenchFlow, Docker: agent and simulator in separate containers	needs a checkout

Output formats#

Output
Trial folder in BenchFlow's trial layout (`config.json`, `result.json`, `verifier/reward.txt`)
Agent trajectory in ATIF v1.7 (`agent/trajectory.json`)
Episode trace (`episode/trace.jsonl`) and video (`artifacts/recording.mp4`)
Task folders in BenchFlow's native format (`task.md`, schema 1.3)