You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Cost-aware LLM router on the cost-quality frontier. Benchmarked on current frontier models (LLMRouterBench: GPT-5, Claude-4, Gemini-2.5-Pro, Qwen3, DeepSeek): matches the top model's quality at 98% lower cost, dominates all 12. Model-agnostic: bring any pool.
Learned LLM router: 95% of GPT-4's quality at 33% of its cost on 36k RouterBench prompts, served by an OpenAI-compatible gateway with fallback, circuit breakers, budgets and metrics
A 421M System-One router for AI agent turns: beats a 27B LLM judge, vLLM Semantic Router and RouteLLM on held-out tier routing, with 0 generated tokens. Pre-registered evals, dataset, checkpoint.
Explainable LLM router: a Growing Neural Gas over prompt embeddings picks the cheapest capable model per prompt. Competitive with UniRoute on RouterBench (11 LLMs, 35k prompts); fully reproducible.