Sakana Fugu
Points and comments are a snapshot, not live.
Sakana Fugu orchestrates multiple LLMs via one API for superior complex task performance.
Sakana Fugu is a multi-agent system that dynamically coordinates a pool of powerful models to tackle complex, multi-step tasks. It offers two models: Fugu (balanced performance and latency) and Fugu Ultra (maximizes answer quality). The system is backed by two ICLR 2026 papers (TRINITY and Conductor) on learned model orchestration, where a coordinator assigns roles (Thinker, Worker, Verifier) or discovers natural-language strategies via reinforcement learning. Fugu achieves frontier-level performance on coding, reasoning, and scientific benchmarks, surpassing models like GPT 5.5 and Opus 4.8 in some benchmarks. It provides flexibility to exclude specific providers for compliance. Available via an OpenAI-compatible API, not yet in EU/EEA due to GDPR compliance.
What commenters are saying
Commenters compare Fugu to OpenRouter's Fusion API. One clarifies: Fusion calls multiple models and synthesizes results, while Fugu uses an orchestrator to decide which model to call, making it more agentic and dynamic. Another commenter notes links to ICLR 2026 papers substantiating quality. Some question the value proposition: "priced the same as frontier models" and ask if it replaces one single-vendor dependency with another. Several commenters suggest ensemble techniques are not new.