How they use Nebius

Chatfuel uses Nebius Token Factory to serve a cascade of Llama 405B models for its chatbot agents. Tasks include constrained text generation, text classification and routing user input to the correct agent. Chatfuel's March 2025 article describes the implementation under the earlier Nebius AI Studio name. Chatfuel also used Nebius Compute Cloud to host solution components and evaluate their quality.

Nebius developed the model cascade and a custom SDK, helping Chatfuel bring the system into production in one month. Nebius developed and tested multiple model versions to balance performance, cost and speed. The account reports improved parsing of customers' free-form agent descriptions, lower LLM costs and a 24% performance improvement over GPT-4o on agent routing. Nebius also reports better response quality and faster interactions for Chatfuel's chatbot agents. Nebius supported training-process optimization using smaller, curated data samples. Evaluation combined an initial Nebius assessment using advanced metrics and basic human evaluation, Chatfuel product-team review and testing by existing customers.