DigestAI news desk
hot OpenAI launches Agents API beta for long-running cloud agents new Meta Reverses Course on Flat Org Structure hot OpenAI board member warns company is not on track to prevent catastrophic AI loss of control new Anthropic reveals Houthis used Claude for ballistic missile development hot Mistral AI raises $3.5B at $24B valuation to push sovereign enterprise AI new ChatGPT Ads Arrive: The Future of AI Search for Brands hot OpenAI Unveils GPT‑6 Astra: Record‑Breaking 3D Rendering, Loop‑Transformer Architecture new Accenture & Google Cloud Form Gemini Enterprise Business Group hot Meta Unveils $100-a-Month AI Agent: Muse new Will AI Really Kill Us All? hot Anthropic builds predictive surveillance system to monitor AI critics and protestors new ChatGPT Defines Upper Class in 5 Years hot OpenAI Launches ChatGPT for Financial Services on GPT-6 Astra new Google Invests €13bn in Finland Data Centers, Secures Nuclear Power hot OpenAI solves Navier-Stokes problem, sparking academic controversy over data use new OpenAI asks Congress if industry-wide AI slowdown would be legal
Generative AI & Models updated

Sakana AI releases Fugu Max and Fugu Ultra v2 for multi-agent orchestration

Sakana AI has introduced two new models, Fugu Max and Fugu Ultra v2, designed to enhance multi-agent orchestration capabilities. Both models utilize a shared learned orchestration architecture but serve different market segments. Fugu Max is optimized for cost-efficiency, routing tasks to lean open-source and specialized models, including NVIDIA Nemotron. It offers competitive pricing at $2 per…

1 source

Key points

  • Sakana AI launched Fugu Max and Fugu Ultra v2, both using a shared learned orchestration architecture.
  • Fugu Max routes tasks to models like NVIDIA Nemotron at $2/$6 per 1M tokens for cost efficiency.
  • Fugu Ultra v2 targets peak capability, scoring 48.3 on Chartography and 74.3 on DeepSWE benchmarks.

In contrast, Fugu Ultra v2 targets peak performance and complex reasoning tasks. The model demonstrates significant improvements in benchmark performance, achieving a score of 48.3 on Chartography and 74.3 on DeepSWE. These results suggest a substantial leap in handling deep software engineering and chart interpretation challenges. By offering both a cost-effective option and a high-performance variant, Sakana AI aims to address diverse enterprise needs within the growing multi-agent ecosystem.

The release highlights the trend toward specialized orchestration layers that manage multiple underlying models rather than relying on a single monolithic LLM. This approach allows developers to balance latency, cost, and accuracy by dynamically selecting the most appropriate underlying model for specific sub-tasks.

Read the original at MarkTechPost · by Asif Razzaq Open source ↗
Topics · follow one to build your own front page
Sakana AINVIDIAFugu MaxFugu Ultra v2Nemotron

The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us.

Comments

via GitHub Discussions

More in Generative AI & Models

All →

Related stories