Sakana AI Launches Fugu Beta: Multi-Agent System Hits SOTA on Three Benchmarks

Sakana AI has launched the public beta of Fugu, a multi-agent orchestration system that dynamically coordinates frontier models and autonomously selects optimal agent combinations per task, reaching state-of-the-art on SWE-Pro, GPQA-Diamond, and ALE-Bench benchmarks.

1 min read|agenticonsult Intelligence

Sakana AI Launches Fugu Beta: Multi-Agent System Hits SOTA on Three Benchmarks

Sakana AI has opened the public beta of Fugu, its multi-agent orchestration platform. Fugu dynamically coordinates frontier models and autonomously selects optimal agent combinations per task, achieving state-of-the-art results on SWE-Pro (software engineering), GPQA-Diamond (graduate-level scientific reasoning), and ALE-Bench (agent language evaluation). The system is available as an OpenAI-compatible API for drop-in integration. Two tiers are offered: Fugu Mini (latency-optimised) and Fugu Ultra (full model pool for deep complex reasoning). Beta access is via an application form at sakana.ai/fugu-beta.

Why It Matters

Fugu's SOTA results across three distinct benchmark categories — coding, science reasoning, and agent evaluation — from a single dynamic orchestration layer positions it as a credible alternative to single-model API calls for complex tasks. The OpenAI-compatible API makes adoption frictionless for any team already building on the OpenAI ecosystem.

Primary source

Sakana AI

This breaking-news item was assembled from the cited primary source with AI assistance. It is intended for rapid situational awareness — refer to the original publication for the definitive statement.

Sakana AI Launches Fugu Beta: Multi-Agent System Hits SOTA on Three Benchmarks

Sakana AI Launches Fugu Beta: Multi-Agent System Hits SOTA on Three Benchmarks

Why It Matters

Live Intel Feed