grail
#81TrainingDecentralized post-training reinforcement learning (RLHF/GRPO) subnet - the alignment and quality layer for Covenant AI's training stack.
Price
24h
7d Change
Market Cap
Emission
Miners
Validators
Competitor Mapping
grail
Decentralized
Market Cap
$30.96M
TAM
$30.00B
Model
Token-incentivized
Team
Decentralized miners
OpenAI (RLHF) + Anthropic (Constitutional AI)
Private
Market Cap
Private
TAM
$30.00B
Model
Revenue-driven
Team
Enterprise
Grail decentralizes the post-training RL step that transforms base models into aligned, capable systems - the same process that made GPT-4 dramatically better than GPT-3. Competes with OpenAI's RLHF and Anthropic's RLAIF pipelines.
Implied Valuation
If grail captures a share of OpenAI (RLHF) + Anthropic (Constitutional AI)'s $30.00B TAM
Bear - 0.1% TAM
$30.00M
implied valuation
vs $30.96M current
Base - 1% TAM
$300.00M
implied valuation
vs $30.96M current
Bull - 5% TAM
$1.50B
implied valuation
vs $30.96M current
Scenarios illustrate potential scale relative to the traditional market. Actual outcomes depend on adoption, tokenomics, and competitive dynamics.
Thesis
Post-training RL is the most valuable step in the AI pipeline - it's what turns raw models into useful products. Grail completes Covenant AI's vertical stack (SN3 pre-training, SN39 compute, SN81 post-training), creating a full decentralized alternative to OpenAI's pipeline. The 211% 30-day token surge after Covenant-72B shows market conviction.
Team
Part of Covenant AI, founded by Sam Dare (same team as SN3 Templar). Covenant AI operates three subnets: Templar (SN3, pre-training), Basilica (SN39, compute), and Grail (SN81, post-training/RL). Team is doxxed.
- Sam Dare - Founder & CEO (Covenant AI) - 8+ years emerging tech, also leads SN3 Templar
Funding
No confirmed VC funding rounds disclosed. Funded through TAO emissions alongside SN3 and SN39.
Traction
SN81 token surged 211% in 30 days (Mar 2026). Running GRPO-based reinforcement learning on validated rollouts. Miners generate rollouts, validators verify proofs and evaluate kernel correctness on-GPU. Pushing into kernel verification for provable training computation.
Recent News
- Mar 2026 - SN81 token surged 211% in 30 days following Covenant-72B announcement
- Mar 2026 - Grail v0 shipped decentralized reinforcement learning capability
- 2026 - Working on kernel verification to prove training computations happened correctly
- 2026 - Covenant AI vertical integration: SN3 + SN39 + SN81 forming 'decentralized OpenAI'
Risk
Entirely dependent on SN3 Templar producing base models worth fine-tuning. If Covenant-72B proves to be a one-time achievement rather than repeatable pipeline, Grail loses its input.