Sakana AI Expands Push Into Japan's Corporate Market With New Fugu Models, Competing on Cost and Performance
TECHWORLD ·
✦ AI Summary
Sakana AI has signed a comprehensive business agreement with Sumitomo Corporation and SCSK, using their corporate customer networks and industrial reach to expand into the corporate market in Japan.
Starting with finance and manufacturing, the three companies will support the linking of each company's work, data and existing systems, and build a framework that backs everything from AI adoption to operations.
Sakana AI unveiled Fugu Max and Fugu Ultra v2, which use orchestration technology that selects and combines multiple AI models for each task.
Sakana AI is expanding its push into Japan's enterprise AI market. The company plans to grow its corporate business by tapping the corporate customer networks of Sumitomo Corporation and SCSK, both of which have touchpoints with enterprise customers in Japan. Its strategy puts orchestration technology, which selects and combines AI models suited to each company's work, at the forefront.
Sakana AI signed a comprehensive business agreement with Sumitomo Corporation and SCSK aimed at expanding the use of AI in Japanese industry. At the heart of the partnership is combining Sakana AI's AI R&D capabilities, SCSK's system integration and deployment capabilities, and Sumitomo Corporation's industrial network.
The three companies will support the linking of each company's work, data and existing systems, and set up a structure that supports everything from AI adoption to real-world application and operations. Through this, they aim to build a system for introducing and running AI in the workplace.
The three companies plan to begin with finance and manufacturing. These sectors are characterized by high requirements for reliability and security. Sumitomo Corporation has about 900 affiliated operating companies and a base of 100,000 corporate customers.
Starting with finance and manufacturing, Sakana AI plans to expand into more applications. The target areas for expansion include finance, manufacturing and cybersecurity. The expansion targets also include cybersecurity, enterprise AI products and services, and social infrastructure.
SCSK's collaboration with Sakana AI starts with applying Sakana AI's orchestration technology to SCSK's Frontier AI business. The two companies will jointly develop a cybersecurity service that supports vulnerability diagnosis and remediation, and they will also jointly develop enterprise AI products. They will also advance cooperation in a direction that builds an AI usage framework that is not tied to any specific model.
This move comes as the use of generative AI by Japanese companies remains relatively lagging behind the US and China. In Japan, demand is growing to move beyond the simple efficiency gains of generative AI. Demand is also rising for expanding its use into actual business and operational systems.
In line with this expansion in the enterprise market, Sakana AI has moved to upgrade its Fugu lineup. On the 11th, the company unveiled 'Fugu Max' and 'Fugu Ultra v2.' Neither product adopts a dedicated single large model approach.
Instead, the two products use an orchestration structure that selects and combines multiple AI models for each task. Fugu Max is focused on cost-effectiveness. Fugu Ultra v2 is focused on maximum performance for complex, multi-step tasks.
Fugu Max broadened the range of open-weight and specialized models it can use, including Nvidia's Nemotron. It is designed to judge the performance needed for each task and then choose a model capable of handling it. In doing so, it reduces calls to expensive frontier models.
Sakana AI said Fugu Max recorded the highest overall score in 6 benchmarks: Terminal Bench 2.1, GPQA Diamond, AA-LCR, GDP.pdf, AutomationBench and SWE-Fish. It also said Fugu Max broke through the cost-performance Pareto frontier of existing single models in 7 of 10 benchmarks.
The price of Fugu Max is USD 2 for input per 1 million tokens and USD 6 for output per 1 million tokens. Sakana AI said Fugu Max's output price is 40% to 60% lower than Sonnet 5, GPT-5.6 Terra and Kimi K3.
By contrast, Fugu Ultra v2 prioritizes performance over cost. It focuses on combining models across multi-step tasks such as multi-step reasoning, autonomous research and software development. Fugu Ultra v2 generates results through model combinations.
In performance evaluations across 8 benchmarks, including GDP.pdf, Chartography, SWE-Fish, DeepSWE and Toolerton, Fugu Ultra v2 took the top or joint top spot in 5 of the 8 and ranked in the top 2 in 7 of the 8. Specific scores cited were 48.3 points for Chartography and 74.3 points for DeepSWE.
In terms of model composition, the company highlighted reduced dependence on specific frontier models. The Fugu Ultra v2 model pool does not include Fable 5, Fable 5.1 or GPT-6 Astra.
Sakana AI said that by using interchangeable combinations of open-weight and specialized models, it can mitigate the impact of a given provider's API price increases, service policy changes or supply interruptions. As a result, companies can avoid dependence on a single model and change model configurations according to purpose and cost.
Sakana AI plans to expand Fugu's applications across corporate operations based on its partnerships with Sumitomo Corporation and SCSK. In the process, it set out an orchestration strategy centered on securing cost, performance and model choice, and the key question is whether that strategy can spread across Japan's corporate market.
Source: TECHWORLD · Kim Seung-gi
Original: https://www.epnc.co.kr/news/articleView.html?idxno=406862
References
This article was produced with the help of an automated content generation algorithm.
Source: TECHWORLD
View originalThis article was summarized and organized by BizCrush based on the original article from TECHWORLD. For exact quotations and full details, please refer to the original article.