SN3 launches on SimpleSwap: Distributed AI model Templar attracts market attention.

Key Insights : Bittensor subnet SN3 (Templar) successfully trained the Covenant-72B (72 billion parameters) large-scale model through a decentralized mechanism, surpassing Meta's LLaMA-2-70B in multiple benchmark tests. This marks the transition of distributed AI training under cryptographic incentives from theory to practice. Following the announcement of this event on March 10, 2026, the price of TAO surged by +40% after a two-day delay, reflecting market perception of an arbitrage window. The news of SN3's listing on SimpleSwap further amplified attention, but current data does not show specific trading details or immediate price reactions; the liquidity injection effect needs to be monitored. (TechFlowPost )

SN3's breakthrough is not isolated, but rather a crucial milestone in the Bittensor ecosystem's transformation from an "incentive protocol" to an "AI infrastructure." Traditional AI training relies on massive data centers and hundreds of millions of dollars in investment, while SN3, using 70+ independent miners and without central coordination, completed training with 1.1 trillion tokens in 6 months. This not only proves that the crypto mechanism can drive serious AI research, but also opens up new boundaries for the open-source community—previously, 70B models were only patented by large companies, but now they can be produced by the community itself.

SN3 Project Background and Core Technologies

Bittensor is a decentralized machine intelligence marketplace, while SN3 (Templar, tplr.ai ) focuses on distributed large-scale model training. Its core innovation is using the TAO token to incentivize gradient contributions : miners upload effective improvements (gradients), and the system automatically scores and settles payments, eliminating the need for trusted third parties.

• Training results : Covenant-72B-Chat was released on March 10, 2026, as an open source (Apache 2.0). The paper can be found on arXiv:2603.08163 , and the HuggingFace model card was publicly validated.
• Scale : 1.1 trillion tokens (≈5.5 million books), 70+ miners participating, maximum of 20 nodes per round, starting from 2025-09-12.
• Key Technologies : | Technology | Breakthrough | Impact | |------|------|------| | SparseLoCo Optimizer | >99% Compression Ratio (146x), GB-level gradient compression to minimum | Solves global node communication bottleneck | | Gauntlet Anti-Cheating | Automatically detects invalid contributions | Ensures incentive fairness | | Communication Overhead | Only 6%, 94% of time is pure training | Comparable to centralized efficiency |

These optimizations have enabled decentralized training to outperform centralized models on mainstream benchmarks for the first time, a feat recognized by the NeurIPS workshop. (TechFlowPost )

Model performance comparison: Surpassing LLaMA-2

Covenant-72B outperforms its meta counterparts in key tests by 4-24 percentage points. While lagging behind state-of-the-art (SOTA) models (such as Qwen2.5-72B's 86.8% MMLU), the key difference lies in the framework: no massive investment, only crypto incentives.

HuggingFace benchmark data, 2026-03-18

Why it matters : This isn't just about piling up parameters, but about a leap in quality. The evolution of decentralized training shows that SN3 jumped from a 6B model in 2022 to 72B, surpassing the centralized benchmark for the first time, with the trend accelerating (12x increase in parameters over 4 years).

Market reaction and the impact of launching SimpleSwap

• TAO Price : After a two-day silence following the announcement, it surged on the third day, and increased by 40% in six days. Reason: Crypto investors view it as an "open-source model," while the AI community ignores crypto, creating a cognitive gap that leads to undervaluation.
• SN3 listing on SimpleSwap : The headline amplifies the event's exposure, but the provided news crawl (as of 09:15 UTC on March 20, 2026) fails to capture specific details such as listing date, initial liquidity, or trading volume. As a non-custodial DEX, SimpleSwap's listing may improve the accessibility of TAO/SN3-related tokens, potentially attracting retail investors. However, the lack of real-time data makes it impossible to quantify the immediate impact (such as 24-hour volume or price fluctuations). Data limitations : This may be a very recent event (<24 hours), or limited to the Chinese community; monitoring CoinGecko/SimpleSwap for real-time updates is recommended.

Evidence that Bittensor is undervalued overall: SN3 successfully verified the protocol's utility but did not immediately boost TAO. Similar to how Bitcoin's value awakening after proving decentralized money takes time.

Risks and Outlook

Positive factors :

• Ecosystem position: Bittensor's multi-subnet competition and SN3 dynamic TAO allocation enhance the survival of the fittest.
• Industry significance: crypto's pioneering contribution to AI technology (HuggingFace/arXiv) has shaken the "monopoly of large companies".

risk :

Outlook : The launch of SimpleSwap on SN3 could catalyze liquidity, and combined with post-halving emissions (3,600 TAO/day), TAO could reach new highs if subnet demand keeps pace. On the bottom line, SN3 proves that "crypto is the hope for AI," but more subnets are needed to validate its sustainability. (TechFlowPost )

Investment perspective : In the short term, focus on trading activity after launch; in the medium term, track the next model iteration. Data is fresh (March 18, 2026), but SimpleSwap details are missing—current TVL/volume is unknown; it is recommended to review within 24 hours.