Dream-1 claims new Pareto frontier for robot control: 16x cheaper, 1.7x faster, 59 vs 48 on RoboDojo
scott_e_reed · x · 2026-10-20
Independent eval by Robocurve ahead of launch puts Dream-1 on a new Pareto frontier for robot control: 16x cheaper and 1.7x faster than the previous best, scoring 59 vs 48 on RoboDojo-RC Tier 1 — the first system above 50. Charts show $0.15 and 1.4 min per task attempt, beating general LLMs (Opus 5.5, Gemini 4 Argon, GPT-6 Astra) and specialized models like MolmoAct2 and GR00T N1.7 across 120 real-world attempts per config. Scott Reed's takeaway: LLMs can plan, but robots still need a cheap, fast, general System 1 to act.
More from Embodied
- Developer's new local agent machine looks like an Xbox 360 — kevinkern · 2026-10-20
- LimX Dynamics TRON 1 humanoid stays upright as kids kick and trip it — burhop · 2026-10-20
- Smart glasses test: Octen search at 60ms, model generation is the bottleneck; Mercury 3 wins — Scobleizer · 2026-10-20
- Cheap Robot Arms Are Turning 3D Printers Into Overnight Factories — Scobleizer · 2026-10-20
- Dream-1 claimed to score 1.3x higher than GPT-6 Astra Ultrafast at 1.7x speed — gharik · 2026-10-20
- Ex-NVIDIA team launches Dream Labs with Dream-1 robot model via public API at $3/hour — gharik · 2026-10-20