Bittensor Subnet Breaks the Decentralized AI Limit, TAO’s March Gain Doubles

The Bittensor Subnetwork 3 (SN3) development team recently released the Covenant-72B model—a large language model (LLM) with 72 billion parameters that has completed permissionless training across 70+ globally distributed nodes. The model scored 67.1 on the general language understanding benchmark MMLU. Driven by this breakthrough, the TAO token price nearly doubled in March.

Covenant-72B: A Decentralized Training Milestone That Rewrites Industry Assumptions

For a long time, the industry has viewed decentralized AI training as a suboptimal option due to its slow speed and difficulty ensuring consistency. The release of Covenant-72B directly challenges this narrative—completing training of a 72 billion-parameter model in 70+ nodes distributed worldwide, in a fully decentralized, permissionless manner, and reaching 67.1 on the MMLU benchmark, approaching the same level as Meta Llama 2 70B.

This is the first time decentralized training has passed a quantitative benchmark test, delivering technical results directly comparable to centralized training, and elevating Bittensor’s positioning from an “experimental decentralized network” to “infrastructure capable of producing practical large-scale AI model outputs,” which in turn has driven the market to reprice the TAO token.

GMAI Index Breakdown: TAO’s Rise Nearly Supports the Entire Market Trend

The GMAI Index’s weekly report on Sunday was 51.26 points, up 48% compared with early February, but still about 84% lower than the historical peak in Q1 2024. To understand this uptrend, you need to look at the index structure: Render (RNDR) and the AI Superintelligence Alliance (ASI) together account for more than 71% of the index weight, TAO accounts for 24.89%, and the combined total of the three is already close to 97%.

Therefore, most of the index’s gains in March essentially reflect the nearly twofold rally of TAO on a per-token basis, rather than a broad-based recovery across the crypto AI sector. As of the time of publication, TAO is about $317, with a market cap slightly above $3 billion, circulating supply of more than 10.7 million tokens, and over 68% already staked—indicating a higher confidence level among long-term holders.

Ecosystem Expansion: Subnetwork Tokens, the Compute Market, and Endorsements from Notable Figures

This Bittensor rally is not limited to TAO itself. The top subnetwork token τemplar (SN3) surged more than 400% over the past month, reaching a market cap of about $130 million. Targon (SN4), a decentralized GPU compute market operated by Manifold Labs, completed a six-figure transaction, providing backend compute power for Dippy AI’s 8.6 million users for AI applications—showing that the commercialization of the subnetworks has moved from concept validation to real revenue.

Public attention from NVIDIA CEO Jensen Huang and well-known investor Chamath Palihapitiya on the Bittensor ecosystem provided narrative endorsement from traditional tech capital for this TAO rally, further expanding the audience’s understanding of the basis for the upward move.

Frequently Asked Questions

What is Bittensor?

Bittensor is a decentralized AI training and inference network that coordinates global distributed nodes contributing compute power through a token incentive mechanism, used to train and run AI models. TAO is its native token; holders can participate in network governance through staking and earn rewards.

What is the core significance of the Covenant-72B model breakthrough?

Covenant-72B is the first large language model trained on decentralized permissionless nodes and achieving MMLU performance at a level comparable to Meta Llama 2 70B. For the first time, it verifies the technical feasibility of decentralized training with a quantitative benchmark test, changing the industry’s existing assumptions about the upper limits of decentralized AI training capability.

Does the GMAI Index’s 48% gain indicate that the overall AI cryptocurrency market is recovering?

Not exactly. The composition of the GMAI Index is highly concentrated; the top three tokens (RNDR, ASI, TAO) together account for nearly 97%. The March index increase is mainly contributed by TAO, rather than reflecting a broader, full recovery in the AI cryptocurrency market. The GMAI Index is still about 84% lower than the peak in Q1 2024.