Breaking

Higgsfield Uses GPT-6 Astra to Auto-Label 13,038 Objects on Warehouse CCTV

September 7, 2026 at 12:58 EDT

Foundation Models
AI Agents
Sales & Business

Higgsfield AI has shown GPT-6 Astra generating 13,038 annotations across 81 frames of parcel-sorting warehouse CCTV, pairing OpenAI’s new flagship with Higgsfield’s tools to turn security-camera video into dense object labels. The company said the same pipeline could track parcel handoffs, monitor worker movement against cart traffic, and flag bottlenecks on the floor.

Sep 7, 2026 · Higgsfield AI

13,038 warehouse labels, no annotation farm

GPT-6 Astra, running with Higgsfield, generated dense object labels across 81 frames of parcel-sorting CCTV — a computer-use demo for logistics, not a shipped labeling product.

13,038

Annotations generated

Warehouse CCTV frames

~161

Labels per frame

86.6%

Vision score, 1st of 53

Object detection, drawn to scale

Low-effort scores on the same vision benchmark. Column height matches the percentage.

53.9%

Field median

82.1%

GPT-6 Astra

14 stacked blocks vs 9 · one block ≈ 6 percentage points

From camera clip to floor ops

1 · CCTV

81 frames of parcel-sorting warehouse video

→

2 · Agent

GPT-6 Astra runs with Higgsfield tools

→

3 · Labels

13,038 object annotations on the clip

→

4 · Ops

Handoffs, worker traffic, bottlenecks

What improved

Dense auto-labels on a live floor. Developers see tight boxes, class IDs, and about 161 marks per frame as a first-pass shortcut on common objects.

What remains

Occlusion, cost, and rare defects. Overlapping parcels, a $0.030-per-sample vision run, and long-tail industrial classes are still open questions. No accuracy audit was published.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.