Is Poke-Aurora state of the art?Pending benchmarks.Announced Oct 30, 2026; no independent results yet.

The details so far

Lab

PokeBot

Announced

Oct 30, 2026

API

Live

Open weights

Closed · Not released

Access

RoboAPI at api.pokebot.com/v1 (OpenAI SDK compatible; sessions plus chat/completions). Free technical preview for non-commercial use; not offered to for-profit third parties. The docs say to pick the model id from /models

Inputs and outputs

Wrist-camera image, an 8-value robot state and an instruction in; action chunks out. PokeBot says English instructions work better

Method

Modulated Residual Coupling; UMI-style post-training with a wrist fisheye camera and end-effector pose control. Technical report not yet published

Robot setups shown

UR7e/UR5e and Franka FR3/Panda with Robotiq 2F-85 grippers, YAM Standard with YAM Gripper, and a floating-base Tianji with YAM Gripper

Supported hardware (docs)

UR5 CB3, UR5e, UR7e or Franka arms; Insta360 Ace Pro 2 or DJI Osmo Action 6 wrist cameras

Tags marked unsure are not stated on the lab's own pages.

What would let us call it

Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.

• Publicly announced with a source
• Usable by the public (API or weights)
• Listed on an independent board we track
• Enough results for a rank in a category
• Holds across two snapshots (crowns only)

What the lab claims

Self-reported. Shown for context; not used in any verdict.

Lab claim: on PokeBot's own evaluation (about 1,000 object instances and about 10,000 instructions), 86.0% on Transfer & Organization, vs 71.3% for a model PokeBot calls Astra, 30.9% for π0.5 and 26.9% for MolmoAct2.

Lab claim: 54.1% on Pose & Shape Alignment (vs 36.3%, 14.8% and 7.9%) and 87.7% on Environment Interaction (vs 67.9%, 32.9% and 23.6%) against the same three baselines.

Sources