← Trends

Open-Vocabulary, Promptable Vision Foundation Models

Vision foundation models shift to text-promptable, open-vocabulary detection, segmentation, and real-time tracking of arbitrary concepts, generalizing perception beyond fixed label sets across images and video and pushing open perception models toward production use.

strengthening · confidence 100 · 0 7d · +11 30d · Medium term (3-9 months) · tracking since June 16, 2026 · updated August 28, 2026

Sign in to get threshold and movement alerts for this trend.

Score history

Daily conviction score, 0 to 100. Higher means the thesis is more strongly corroborated.

Aug 27 · 96Aug 28 · 100

Now 100 · +4 since Aug 27 · ranged 96 to 100

Showing the last few days. Unlock full score history.

Why the conviction moved

  • Aug 28
    Strengthened +5

    Berkeley Lab's SYNAPS-I project cut beamline image analysis from about a month to roughly 15 minutes by running Meta's SAM 3 and DINOv3 across 300 Nvidia A100 accelerators. This is open-vocabulary perception models producing a measured production result on scientific imagery they were never trained for, the generalization claim the thesis rests on.

  • Aug 25
    Strengthened +6

    Berkeley Lab put Meta's Segment Anything 3 and DINOv3 on 300 Nvidia A100s at NERSC to segment X-ray and neutron imaging data in real time, cutting beamline analysis from a month to fifteen minutes. Promptable general-purpose segmentation transferred to scientific imaging with no domain-specific label set, which is the open-vocabulary claim tested outside consumer imagery.

Showing the last 2 days. Unlock the full record.

Source trail

  • Supporting · August 28, 2026

    Berkeley Lab Cuts Beamline Image Analysis From a Month to About 15 Minutes Using Meta's Open Vision Models

    Berkeley Lab's SYNAPS-I project cut beamline image analysis from about a month to roughly 15 minutes by running Meta's SAM 3 and DINOv3 across 300 Nvidia A100 accelerators. This is open-vocabulary perception models producing a measured production result on scientific imagery they were never trained for, the generalization claim the thesis rests on.

    Meta AI
  • Supporting · August 25, 2026

    Meta's Open Vision Models Cut Beamline Analysis From a Month to Fifteen Minutes at Berkeley Lab

    Berkeley Lab put Meta's Segment Anything 3 and DINOv3 on 300 Nvidia A100s at NERSC to segment X-ray and neutron imaging data in real time, cutting beamline analysis from a month to fifteen minutes. Promptable general-purpose segmentation transferred to scientific imaging with no domain-specific label set, which is the open-vocabulary claim tested outside consumer imagery.

    Meta AI (Genesis Mission)

Unlock full source trail, score history, and daily updates.

54 more sources in the full trail.

Unlock Trends

Affected regions & assets

Assets1 assetUnlock Trends

Townsquare

Argue the thesis in Townsquare.