ScaleVLA released
The story so far
DeepSeek has released the ScaleVLA model, extending the Visual Language Action Model to 552 billion parameters, connecting the 552B DeepSeek-V4.1-Flash backbone with trainable action DiT. Compared with VLMs from 2B to 552B in shared training settings, ScaleVLA has improved in sample efficiency, generalization ability, and precise operation: 10 demonstrations surpass 2B backbone with 50 training samples (87.8% vs. 85.6%), LIBERO-PRO success rate reaches 45.1% (highest 31.3% for small models), and charger insertion success rate increases from 4/10 to 9/10. Despite the large parameter size, ScaleVLA's median latency per action block is 216 ms.
AI-written from the reports · Updated 17 minutes ago
Coverage timeline
Follow the coverage to see every side of the story.
- Hugging Face Daily PapersScaleVLA: Scaling Vision-Language-Action Models to 552 Billion Parameters
ScaleVLA connects the 552B DeepSeek-V4.1-Flash backbone with trainable action DiT, comparing VLMs from 2B to 552B under shared training settings. 10 demonstrations surpass 2B backbone with 50 training samples (87.8% vs. 85.6%), LIBERO-PRO success rate reaches 45.1% (highest 31.3% for small models), charger insertion success rate improves from 4/10 to 9/10.
Heat over time
Not enough data yet to draw a trend.