On September 11, Sakana AI announced "Fugu Ultra v2" and "Fugu Max," the latest versions of "Sakana Fugu," an AI system that combines multiple AI models to perform tasks.

The company stated that Fugu Ultra v2 showed results surpassing the scores of models such as OpenAI's "GPT-6 Astra" and Anthropic's "Claude Fable 5.1" in benchmarks such as "DeepSWE," which measures coding performance. According to the company, Fugu Ultra v2 achieved these high scores without including those models as collaboration partners.

Meanwhile, Fugu Max is a model specializing in cost-performance. It newly includes open-weight models such as NVIDIA's "Nemotron" series as collaboration partners; the company claims this has reduced output costs by 40% to 60% compared to other companies' "GPT-5.6 Terra" and "Claude Sonnet 5."

Both models became available on the same day through an OpenAI-compatible API. The API pricing for Fugu Ultra v2 is $5 per million input tokens ($0.5 for cached input) and $30 for output; for Fugu Max, it is $2 for input ($0.25 for cached input) and $6 for output.

Source: