Qwen3.8-Flash-Next Reaches Closed-Model Robotics Performance via Open-Source uni-robot-agent Harness

·2026.11.04 02:03

Key point

The open-source uni-robot-agent harness enables Qwen3.8-Flash-Next to reach the performance level of some closed-model robotic agents on real-world manipulation tasks without model weight changes.

Details

uni-robot-agent is an open-source harness by Qwen developers from Tsinghua and UC Berkeley that connects general-purpose LLMs to robotic systems via tool calls without fine-tuning.
Qwen3.8-Flash-Next reached the performance level of some previously reported closed-model robotic agent systems on 19 real-world robot manipulation tasks.
Average API cost for Qwen3.8-Flash-Next was $1.09 per task, compared to $21.29 for GPT-6 Astra.
Without robot-specific fine-tuning, Qwen achieved a slightly higher score than the fine-tuned pi_0.5.
Local deployment tests showed the harness score with Qwen3.8-Flash-Next rising from 1.90 to 31.39, surpassing the RoboDojo Team's score with GPT-6 Astra.

robotics open-source qwen llm-agents uc-berkeley tsinghua

Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.