RunwaySpecialist modelOpen weights
Praxis-1
n/ano search or usage data
Summary
Praxis-1 is Runway’s world action model for robotics. It predicts how a scene and a robot will move together. Alongside the base model, Runway offers weights for a single DROID arm and two YAM arms. Both robot configurations also have weights for NVFP4 inference.
The weights and official inference code are publicly available. Hosted inference is available through Runway Dev and Reactor.
On this page
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 2 published benchmark results
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
Praxis-1 Figure 2, web video from scratch, 1000 h (Show measurement, test conditions, and source)
- Source value
- 16.1
- Metric
- Final robot-object placement error in cm
- Unit
- No unit provided
- Category
- robotics policy evaluation
- Direction
- Lower is better
- Sample
- 93
- Source type
- vendor-reported
- Evaluator
- Runway Research, Introducing Praxis-1, Figure 2
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Vendor-reported result. Runway's Praxis-1 introduction, Figure 2, endpoint at 10^3 hours of video pretraining followed by finetuning; condition: web video, from scratch. The chart measures final placement error; lower is better.
- Context
- Runway places the figure on the Praxis-1 introduction page and the adjacent text attributes the measured policy-performance scaling to the Praxis-1 approach. The page does not identify immutable weights or a checkpoint revision, robot embodiment, hardware, task prompts, finetuning recipe, or per-condition raw results. The 93 evaluation pairs and ±1 SEM are figure-level reporting, not 93 independent runs per condition. SEM is not a 95% confidence interval. This is vendor-reported, not an independent reproduction.
Praxis-1 Figure 2, teleop robot video from scratch, 1000 h (Show measurement, test conditions, and source)
- Source value
- 16
- Metric
- Final robot-object placement error in cm
- Unit
- No unit provided
- Category
- robotics policy evaluation
- Direction
- Lower is better
- Sample
- 93
- Source type
- vendor-reported
- Evaluator
- Runway Research, Introducing Praxis-1, Figure 2
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Vendor-reported result. Runway's Praxis-1 introduction, Figure 2, endpoint at 10^3 hours of video pretraining followed by finetuning; condition: teleoperated robot video, from scratch. The chart measures final placement error; lower is better.
- Context
- Runway places the figure on the Praxis-1 introduction page and the adjacent text attributes the measured policy-performance scaling to the Praxis-1 approach. The page does not identify immutable weights or a checkpoint revision, robot embodiment, hardware, task prompts, finetuning recipe, or per-condition raw results. The 93 evaluation pairs and ±1 SEM are figure-level reporting, not 93 independent runs per condition. SEM is not a 95% confidence interval. This is vendor-reported, not an independent reproduction.
Similar models
More models of the same kind as Praxis-1, chosen by model line, developer, size, price and release date.
- GWM Worlds 2Runway · same developer
- MetaEncoder v1Meta · similar size
- AstaBrief-8BAllen Institute for AI · released around the same time
- pplx-embed-v2-context-9bPerplexity · released around the same time
- pplx-embed-v2-late-9bPerplexity · similar size
- FLUX 3 ActionBlack Forest Labs · released around the same time
- Molmo2-4B-tbAi2 · released around the same time
- Qwen-Drive-1.0-4BQwen · released around the same time
All specifications
Architecture
- Model class
- Mixture-of-experts diffusion transformer with 27.2 billion parameters in the full robotics model.
- Source: Model class
- NVFP4 variants
- runwayml/praxis-1-droid-nvfp4 and runwayml/praxis-1-yam-nvfp4. The weights are stored in BF16. At load time, the inference code can pack the routed experts as NVFP4.
- Source: NVFP4 variants
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
- Access
- Public weights and official inference code on Hugging Face and GitHub. Hosted inference is available through Runway Dev and Reactor.
- Source: Access
- Shutdown
- None announced
- no source
Further details
- Weight variants
- runwayml/praxis-1-base: general base model without robot heads. runwayml/praxis-1-droid: single DROID robot arm. runwayml/praxis-1-yam: two YAM arms for bimanual tasks.
- Source: Weight variants
Sources and data date
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.
Without vendor claims and historical or disputed results; those stay labelled in the single values.
| Type | Evidence and data date |
|---|---|
| Research status | Research date October 4, 2026. 1 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not found in the inspected sources | exact-checkpoint-benchmark Original wording The publication's 0.95 correlation and 16.1/16.0 cm placement errors belong to its pretraining or policy study, not a named Praxis-1 checkpoint. |
| Not found in the inspected sources | exact-model-quality-benchmark Note Checked official introduction contains no exact Praxis-1 benchmark score; unrelated system or product metrics are not transferred. |
| Not published | praxis-1-external-task-score Note The checked official research page contains no numeric result for an identified external benchmark that can be attributed to Praxis-1; demonstrations are not treated as benchmark scores. |
| Additional source | Runway Praxis-1 announcement (retrieved October 3, 2026; October 15, 2026; October 16, 2026; October 18, 2026) · Runway Praxis-1 announcement · Editorial description reviewed November 12, 2026 |
| Additional source | Runway: Praxis-1 is now available (retrieved November 12, 2026) · Runway: Praxis-1 is now available · Editorial description reviewed November 12, 2026 |
| Additional source | Runway Praxis-1 technical report (retrieved November 12, 2026) |
| Additional source | Runway Praxis-1 Open Source repository (retrieved November 12, 2026) · Runway Praxis-1 Open Source repository · Editorial description reviewed November 12, 2026 |
| Additional source | Runway Praxis-1 model collection (retrieved November 12, 2026) · Runway Praxis-1 model collection · Editorial description reviewed November 12, 2026 |