Table of Contents
Verathos has expanded its AI inference network to nine live models, giving developers access to a wider range of performance, context, and pricing options through a single API.
The current lineup spans lightweight models optimized for speed through larger models designed for more demanding workloads. Kvasir 1.0, Verathos’ network mixture-of-experts model, routes requests across three miners and supports context windows of up to 512k tokens. GLM-5.2 extends that to as much as 1.024 million tokens, while models including DeepSeek-V4-Flash, Qwen3.6, Qwen3.8, and Ornith-1.5 provide additional tradeoffs across cost and performance.
When using these models on Verathos, every response is accompanied by a cryptographic proof. This means that you don’t just have to trust that the AI ran your request correctly. Each response can be checked to confirm the work was actually done as claimed.
That verification layer is central to Verathos’ broader goal of making decentralized AI inference easier to consume through a conventional API while retaining the trust-minimized properties of Bittensor.
Antonio Verrico
The platform is also preparing to bring additional models developed elsewhere in the Bittensor ecosystem onto the network. Its interface currently lists Albedo, Affine, Teutonic, and IOTA as upcoming integrations, potentially turning Verathos into a broader distribution layer for models produced across multiple subnets.
