Primary use stated or inferred cautiously from official documentation.
Praxis-1 DROID
Praxis-1 policy fine-tuned for a Franka arm with a Robotiq gripper and exterior and wrist cameras. It predicts video and actions jointly in 1.6-second chunks. BF16 weights are available under Apache 2.0.
The essential
Maximum input capacity when published by the source.
Documented maximum generation limit.
Verified availability channels.
Primary functional family and verified specialties.
Declared terms for API access, model weights, or self-hosting.
Limits and integration
| API ID | runwayml/praxis-1-droid |
|---|---|
| Model type | Robotics · Video · Multimodal |
| Access model | Downloadable weights |
| License | Apache-2.0 · dependencies under their own licenses |
| Deployment | Self-hosted |
| Release | 11 NOV 2026 |
| Knowledge cutoff | Not published |
| Entrance | Text · Image · Robot state |
| Exit | Video · Robot actions |
| Context window | Instruction, cameras and robot state |
| maximum output | 24 steps at 15 Hz; 10-dimensional actions |
| Reasoning | Not published |
| Published tools | |
| Structured Outings | Not published |
| Batch processing | Not published |
| Prompt cache | Not published |
| Fine-tuning | Not published |
| Verified platforms | Hugging Face · LeRobot |
Single-arm manipulation and evaluation in the DROID environment. Actions include position, rotation and gripper control. The LeRobot distribution represents the same policy, not a separate model.
Reference BF16 inference requires an 80 GB GPU. Weights occupy 54.4 GB, which is not total memory usage. Wan2.2 VAE and Qwen3.5-2B are distributed separately under their own licenses. Long tasks and similar-looking objects remain limitations.
What it can do
Orientation
Single-arm manipulation and evaluation in the DROID environment. Actions include position, rotation and gripper control. The LeRobot distribution represents the same policy, not a separate model.
Context
Instruction, cameras and robot state · 24 steps at 15 Hz; 10-dimensional actions
Tools and integration
Not published
Access
Hugging Face · Runway code
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | Not published | Standard API rate |
| Cached input | Not published | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | Not published | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
No hosted API rate has been verified for this checkpoint. Self-hosting consumes infrastructure. BF16, NVFP4 and the LeRobot layout must be assessed separately; their requirements and results are not interchangeable.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
How to read the results
Runway reports 46.2% success on RoboLab-120 after fine-tuning Praxis-1 on DROID, across 1,200 episodes and ten trials per task. This simulation evaluation applies to the DROID checkpoint, not Base, YAM or every real-world setup.
| Benchmark | Result | Metric | Source |
|---|---|---|---|
| RoboLab-120 | 46,2 % | Simulation success | View source ↗ |
Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.
Chronology
runwayml/praxis-1-droid
Public weight release; the initial September announcement was not general availability.