Bittensor shipped a network upgrade today that slashes root-claiming fees and lets validators run copy-tradable baskets, effectively turning root validators into yield asset managers. Hippius (SN75) launched its Android app with encrypted backup, and a Harvard team released a year-long dataset of Chutes (SN64) inference traffic covering 6.12 billion requests across 9,174 models. Elsewhere, NEAR AI landed on say_gm_, SOMA gets this week's Novelty Search spotlight, and a new decentralized math/code model debuted.
Full coverage
4 reports
01 / DAILY
Root Claiming Gets Cheaper, Validators Get New Powers
Today's Bittensor upgrade cut root-claiming fees from roughly 0.06 TAO to about 0.008 TAO per claim, according to community reports. It also introduced copy-trading of validator baskets, a mechanism that lets root validators function as de facto yield asset managers. Wallet support hasn't fully caught up: the Crucible wallet's Chrome extension failed to load network fees or confirm swaps of root rewards, an issue attributed to the extension not yet supporting the new upgrade.
02 / DAILY
Hippius Brings Encrypted Backup to Android
Hippius, running on SN75, launched its mobile app on Android via Google Play. The app automatically backs up photos and files with end-to-end encryption before upload, sharing the same account and storage as the existing desktop and web product. An iOS version is coming next.
03 / DAILY
Harvard Releases a Year of Chutes Inference Data
A Harvard research team, working with Chutes (SN64), publicly released a dataset covering a full year of real-world LLM inference on the subnet: 6.12 billion requests across 9,174 models. Headline findings include AI agents processing more tokens while producing shorter answers, faster and cheaper inference through reuse of previously processed information, and reduced compute waste from smarter GPU routing.
04 / QUICK HIT
NEAR AI's confidential models are now available through say_gm_, described as an OpenRouter-style gateway for private AI on Bittensor. SOMA (SN114) is the subject of this Thursday's Novelty Search feature, building a decentralized marketplace for AI services starting with context compression, where miners submit compression algorithms scored by validators.
Reliquary-4B, a 4B-parameter math and code model, launched as a decentralized post-trained model built via reinforcement learning, with independent miners generating rollouts and validators verifying them.
A free 'Albedo Agent API' launched, offering miners 10x standard quota for use with coding agent tools such as Claude Code, Codex, Copilot, and Cursor.
