Fine-tuned Dream-1 world action model jumps 159% on real-robot benchmark at 1/100th the cost
paigeinsf · x · 2026-11-19
Fine-tuning Dream-1 (a world action model from Dream Labs) on 500 real-robot demos across 10 tasks lifted its RoboDojo-Real score from 8.5 to 22 (+159%), beating fine-tuned π0.5 (20.6) but still trailing LLM methods overall:
- Fable 5.2 zero-shot still leads at 38.5, winning 7/10 tasks on memory, sorting, and precision
- Dream-1 is 100x cheaper and 5x faster: $0.10 and 3 min per episode vs $10 and 15 min for Fable 5.2
- Dream-1 excels at repetitive physical routines: box packing 54.5 vs 39, peg fitting 35 vs 31
Takeaway: LLMs know what to do; WAMs know how to do it.
More from Embodied
- SE3 Labs Releases A Wider Dream Eval, Opens 500-Episode Robot Demo Dataset on Hugging Face — paigeinsf · 2026-11-19
- LLMs zero-shot beat fine-tuned world action models on 7 of 10 robot tasks — paigeinsf · 2026-11-19
- Memory and Reasoning Remain LLM Territory: Fable Beats Dream-1 64 to 9 on Cup Task — paigeinsf · 2026-11-19
- The Secret Wasn't a Bigger Model: 500 Randomized Demos Lift Dream-1 Score 159% — paigeinsf · 2026-11-19
- Fine-tuned World Action Model Dream-1 Jumps 159%, Beats Frontier LLMs on Robot Tasks — paigeinsf · 2026-11-19
- Robotics founder: "sim2real is BS" — labs buying data is a flash in the pan — Scobleizer · 2026-11-19