Score

Score is a decentralized computer vision platform built on Bittensor Subnet 44. Its core business is turning raw video into structured, actionable data through an open marketplace where miners compete to build small, specialized vision skills (under 30MB) that run on edge devices. The platform's narrative centers on "programmable vision": instead of training one giant model, it composes tiny, benchmarked skills like "detect person" or "track ball" into custom pipelines.

The main pain point Score addresses is the high cost and inefficiency of traditional video analytics, which relies heavily on manual annotation. Score's Turbo Vision system uses vision-language models to auto-generate pseudo-ground-truth labels, cutting annotation costs by 10-100x. Another problem is that general-purpose models are too large and slow for real-time edge deployment; Score's specialist skills are roughly 400x more efficient than frontier alternatives while matching accuracy.

In the past six months (March-September 2026), Score launched Satori 1.0, a 2B-parameter vision-language model distilled from frontier models that runs on-camera. It also opened Manako as a public platform with SDK access, allowing developers to deploy composable vision pipelines via simple APIs. Commercial deployments expanded into petrol stations (AVIA), agriculture (Two-a-Day), and vehicle wash infrastructure (Lavance). Score Studio, scheduled for September 16, will serve as the front end for training, evaluating, and deploying models.

One notable negative event: in early 2025, a cruise ship incident involving a passenger death raised questions about the reliability of AI vision systems in safety-critical scenarios. Score framed this as a system architecture problem rather than a tuning issue, arguing that specialist skills with verifiable performance certificates are the correct approach—but the incident highlighted the inherent risks of relying on automated perception for life-safety decisions.