Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

ACE-Brain-0.6 leads a reported visual-grounding average

AI summary

A visual-grounding guide reports that ACE-Brain-0.6 averaged 75.41 across seven benchmarks, compared with 73.68 for GPT-6 Astra. Its RobotGrounding score was 94.13 versus 91.10, while GPT-6 Astra led on PixMo-Count, RefSpatial and PointArena. The guide also presents the 4B GroundingPI as a lightweight option, with a reported 73.68% average across 34 benchmarks. It cautions that the ACE-Brain-0.6 and GroundingPI scores were self-reported and not independently reproduced, while averages covering different benchmark sets may not be directly comparable.

In-depth insight

The comparison offers scenario-specific guidance for choosing between specialized open-weight models and a proprietary general model. However, its overall state-of-the-art claim remains provisional because the central averages have not been independently reproduced.

核心结论:截至 2026 年 11 月,「靠一句话在图片中定位目标」最准的专用模型是 ACE-Brain-0.6(9B,基于 Qwen3.5-9B 微调)——它在 7 个公开视觉定位 benchmark 上平均 75.41,超过了 OpenAI 的 GPT-6 Astra(73.68),成为当前视觉定位(visual grounding)的 SOTA(state-of-the-art,最先进水平)。但要注意:它是在「空间定位」这类通用感知任务上整体领先,在 PixMo-Count(计数)、RefSpatial(空间关系)、PointArena(点定位)三个子项上仍输给 GPT-6 Astra。所以「最准」要分场景:追求极致通用定位能力选闭源的 GPT-6 Astra,追求可私有化部署、空间定位精度选开源的 ACE-Brain-0.6 或 GroundingPI(4B,34 benchmark 平均 73.68%)。本专栏在视觉定位技术能力边界与选型指南里讲过视觉定位的技术全景,本文更新到 2026 年底的最强选手。

#ACE-Brain-0.6#GPT-6 Astra#GroundingPI#Qwen3.5-9B#ACE Robotics#OpenAI#XPENG

Original · 11 min readRead the original
Group chat

No comments yet. Start the conversation.