13:I[1537,["972","static/chunks/972-a152a4e21020df84.js","447","static/chunks/447-7566e7c6695405bc.js","192","static/chunks/192-42f0cdc837d8115b.js","908","static/chunks/908-5dd9a376a6c3dac4.js","726","static/chunks/726-ab787a8dbb36b15f.js","462","static/chunks/462-11dbc1707a46f572.js","325","static/chunks/325-75dc341894f781b0.js","479","static/chunks/app/market/%5BsubnetId%5D/page-2f43fd2a743d238e.js"],"default"] 14:T5e4,

Grail is a decentralized post-training platform for large language models, built on the Bittensor network as Subnet 81. It enables permissionless reinforcement learning (RL) by coordinating miners, validators, and trainers to collectively fine-tune AI models using cryptographically verifiable inference. Grail implements the GRAIL protocol (Guaranteed Rollout Authenticity via Inference Ledger), which ensures that every rollout produced during RL training is tied to a specific model and input, and can be independently verified on-chain.

As the post-training layer of the Covenant AI ecosystem, Subnet 81 completes the decentralized AI development pipeline alongside Templar (SN3) for pre-training and Basilica (SN39) for GPU compute. Together, these three subnets form a fully open and incentive-aligned stack for building production-grade AI models without relying on centralized labs or infrastructure.

15:T566,

Grail op

Not Available