Score Is Expanding From Computer Vision Into Vision-Language Models
One of Kraken's seven listed subnets is betting the same competitive-training mechanism that produced good detection and segmentation models generalizes to bigger multimodal ones.
BitExplorer · Jul 26, 2026
Score built its name on decentralized computer-vision competitions — detection, segmentation, and similar narrow tasks scored against real benchmarks. It just announced a scope expansion: extending its miner competitions toward training full vision-language models, not just single-purpose vision skills.
It's a real bet, not a guaranteed one. Score is taking a mechanism that's worked at a narrower scope and betting it scales to something considerably more complex. Vision-language models have to do more than label what's in an image — they have to reason about it in natural language.
Score is also one of the seven subnets Kraken picked for its subnet alpha token listing roadmap — real external validation of the subnet's standing, independent of this specific expansion.
Check the Score project page →
Source: @webuildscore · Jun 26, 2026