Cascade (SN91) Outperforms Toto2. Here’s What the Founders Say About Where the Subnet Is Going

Cascade (SN91) has reached 98.4% of the performance of Datadog’s Toto2 time-series forecasting model while using approximately 85% less training data.

The subnet trained its model from scratch on 500 billion tokens, compared with Toto2’s 3.4 trillion. Although it has not achieved full parity across all benchmarks, Cascade reports that it has already outperformed Toto2 on major forecasting benchmarks.

Cascade (SN91) develops AI models that predict future events using historical data, including cryptocurrency prices, weather patterns, and business demand.

However, training these models can be difficult because much of the required data is privately held or difficult to access. To address this, Cascade rewards miners for developing synthetic data generators, with their contributions judged by how much they improve forecasting accuracy.

Speaking on the Shizzy Unchained, the team explained that the subnet adopted Datadog’s model architecture to test whether better training data alone could deliver comparable performance. The results suggest that carefully generated synthetic data can significantly reduce the resources needed to train competitive forecasting models.

Cascade is also building products around its technology. Its Ephemeris API provides access to forecasting models, while a separate evaluation platform uses fresh daily data to compare their performance.

The team has combined six models into an ensemble that it says currently leads its evaluation leaderboard. Potential applications include financial forecasting, inventory planning, and business operations.

The subnet has also developed experimental trading tools, including an agent harness and a beta platform for forecasting TAO and alpha prices. Its initial trading experiments on Astrid Arena (SN127) showed promising early results but ultimately ended in losses.

Cascade’s current model has approximately 4 million parameters, with plans to scale to 22 million and eventually billions. The team is also exploring enterprise forecasting services and intends to use potential profits from mining other subnets to fund SN91 buybacks.

With its model approaching Toto2’s performance using a fraction of the training data, Cascade’s next challenge is to achieve full benchmark parity and demonstrate that its forecasting technology can deliver reliable results in commercial applications.

Check how you can use Cascade for your personal needs: