TLDR;

• Network Governance: BIT-0016 proposes reintroducing subnet deregistration for low-demand subnets once the 256 limit is reached, with a six-month protection period for new subnets.

• Institutional Access: xTAO lists on the TSX, Safello to announces TAO-backed ETPs in Europe, and TAO Synergies files to raise up to $100M for growth and acquisitions.

• Ecosystem Tools & Marketplaces: Taostats launches a Subnet Marketplace; Ridges debuts OpenMine (Google login mining) and Pilot (real-time miner analytics); multiple subnets ship major upgrades including Gradients’ Qwen 3 8B model.

• Ecosystem Expansion: New launches include Dippy Studio for decentralized media inference, Masa’s Unified Search API, VidaIio’s video compression tech, and Macrocosmos’ Apex 3.0 incentive redesign.

• Media & Community: Podcasts and events highlight TAO adoption trends, project milestones, and technical deep dives, while BT Labs announces the Aug. 30 Brunchtensor meetup in Brooklyn.

Highlights of the Week

⚡️ A new proposal, BIT-0016, seeks to reintroduce subnet deregistration, removing low-demand subnets (ADR < 1) once the 256-subnet limit is reached, with new subnets protected for six months before eligibility.

⚡️ The TSX Venture Exchange has officially welcomed xTAO (TSXV:XTAO.U) to the market, while Safello announced plans to list TAO-backed exchange-traded products across five European exchanges, providing non-crypto-native investors and institutions with tax-advantaged access to TAO exposure. Meanwhile, TAO Synergies (TAOX) has filed a prospectus to offer up to $100 million in various securities over time, giving the company flexibility to raise capital for future growth or acquisitions without committing to a single issuance timeline.

⚡️Taostats launched a Subnet Marketplace, allowing verified subnet owners to list subnets for sale with stated prices, reasons for selling, and optional alpha inclusion. Buyers can make offers for a 0.1τ fee, with escrow-assisted transfers and a public, transparent listing process.

Novelty Search

This week’s Novelty Search featured Wandering Weights from Gradients (SN56). Here were the key takeaways:

• Gradients as Best-in-Class Post-Training Platform: Supports Instruct, DPO, and new GRPO methods for LLMs and diffusion models, consistently outperforming major platforms like Hugging Face and Databricks on diverse datasets.

• Major Feature Upgrades: Added larger model support (up to 70B), custom datasets, automated image captioning for diffusion, and flexible reward functions for GRPO.

• Open-Source Tournaments: Miners now submit full training scripts, enabling transparency, collaboration, and aggregation of best techniques, while aligning with Bittensor’s open-intelligence ethos.

• Enterprise & Revenue Focus: Open sourcing addresses client trust concerns, enabling direct customer servicing and positioning Gradients as the most cost-effective fine-tuning option.

• Future Roadmap: Plans include integrating with compute subnets for efficiency, expanding to new modalities (e.g., video, object detection), and cementing leadership with the best 8B model on the market.

Subnet Updates, Ecosystem Developments, & Media ⬇️