Sakana AI launches Fugu Max and Ultra v2
Sakana AI has updated its Fugu multi-agent orchestration platform with the release of Fugu Max and Fugu Ultra v2 across cost efficiency and peak capability axes. Fugu Max routes subtasks to specialized models like NVIDIA Nemotron for $2/M input and $6/M output tokens, while Fugu Ultra v2 targets demanding benchmarks through an OpenAI-compatible API without relying on closed frontier models.
The AI industry is shifting from brute-force foundation model scaling toward intelligent orchestration layers, where compounding collective architectures deliver superior Pareto efficiency and structural resilience. Dynamically routing task steps to right-sized models cuts token costs by 40–60% while matching or exceeding the performance of monolithic frontier LLMs. Decoupling agentic pipelines from individual closed models protects developers from API revocations, vendor lock-in, and sudden cost fluctuations. Furthermore, specialized open-source models such as NVIDIA Nemotron gain outsized collective utility when coordinated in task-optimized agent topologies rather than evaluated in isolation.
DISCOVERED
1h ago
2026-09-11
PUBLISHED
1h ago
2026-09-11
RELEVANCE
AUTHOR
omarsar0