Table of Contents
Score Studio is now live, giving builders a self-serve workspace for creating and deploying computer-vision systems backed by Bittensor Subnet 44.
Score announced the launch on Sept. 16, saying the product brings annotation, training, evaluation, and deployment into one workflow. The product is the builder-facing side of Score’s broader computer-vision stack, alongside enterprise deployments through Manako and research bounties through Vision Lab.
The launch turns Score’s subnet activity into a more conventional product surface, with Score’s computer-vision capabilities now accessible as a tool that independent engineers, small teams, and agents can use directly.
Score Studio launch video
Score Studio Packages the Computer-Vision Workflow
Score Studio is a connected workspace for generating or uploading visual data, labeling it, training model candidates, evaluating performance, composing workflows, deploying approved systems, and monitoring production behavior.
Building a production vision system often requires stitching together separate tools for data management, annotation, model training, evaluation, deployment, and observability. Score Studio collapses those steps into one workflow, where users can start with existing data and models or bring only the problem they need solved.
The workflow operates in seven stages: data, annotate, train, evaluate, build, deploy, and monitor. It’s essentially a visual workflow canvas for connecting inputs, models, vision-language models, transformations, decisions, providers, and outputs. Users can auto-generate and label data, train and evaluate candidate models against defined criteria, deploy approved revisions to connected infrastructure, and feed production evidence back into the next iteration.
In Score’s framing, the goal goes beyond training a model once to keeping the evidence behind each release visible as the system improves.
How Score Studio Connects to Bittensor SN44
The Bittensor-specific part of the launch is the connection between Score Studio and Subnet 44, the Score subnet.
Score Studio is intended as the builder and independent engineer path for using Score models, while Manako serves as the enterprise deployment layer for customers that want computer vision running on existing camera infrastructure. The product can be thought of as an alternative to Roboflow for small teams and builders.
There’s a two-track structure for Score’s model work. A public track is framed around open models, while a private track is framed around commercial models whose evaluation results remain visible through Score’s console. That split lets the subnet support open development while also serving customer-specific use cases that cannot be handled fully in public.
Score’s product site also advertises an agent-facing surface, describing Score Studio as “the computer vision layer for agents” and listing an MCP endpoint for integrating the service into agent workflows.
Also relevant to Score’s subnet is Vision Lab, a mechanism where a customer can describe a vision task, fund a bounty, and have Bittensor miners compete to produce the winning model.
Building the Future of Specialized Vision Systems
Specialized vision problems often fail not because teams cannot access a model, but because they cannot maintain a reliable loop around examples, edge cases, evaluation, deployment, and retraining. Score Studio is trying to turn that loop into the product; already, Score’s specialized models are far cheaper than running comparable full-match sports analysis through large frontier models.
Score Studio now gives builders a workspace for developing and evaluating vision systems, while Manako deploys specialized models into existing camera fleets, already having notched deployments across fuel retail, sports, vehicle washing, physical security, manufacturing, and professional services. Those deployments included more than 120 AVIA sites, 60 Shell and Eni franchisee locations, sports-related work with Reading FC, Eyeball, and Mettle, vehicle-washing deployments with Lavance, and a PwC France and Maghreb alliance or pilot pipeline.
Moving forward, Score plans to bring subnet-trained vision-language models into Score Studio and, over the longer term, pursue an open-source world model for rare or dangerous environments.
