Sakana AI

Researchers from Sakana AI and UC Berkeley introduced Multi-Agent Self-Supervision (MASS), in which a single model solves tasks, judges results and optimizes workflows. Applied to Qwen3.6-27B, the win rate against the base model reached 69.9% after two rounds. But the self-evaluator sometimes accepted flawed or manipulated solutions as successes and fed them into the next training cycle.

Sakana AI has released Fugu Ultra v2, described as a learned multi-agent orchestration system rather than a single monolithic model. It offers a 1M-token context window, configurable reasoning effort, and pricing of $5 per 1M input tokens and $30 per 1M output tokens.

Sakana AI has released Fugu Max, the cost-performance tier of its Fugu family, which uses a trained language model to route tasks across a pool of open-weights and specialized models rather than functioning as a single monolithic network. The model offers a 1 million token context window at a flat $2 input / $6 output per million tokens.

Sakana AI has released Namazu, a reasoning model built on Kimi K2.6 and fine-tuned for Japanese language and business contexts. The model offers a 262K token context window at $0.95 per 1M input tokens and $4 per 1M output tokens.

Sakana AI has released Fugu Ultra v1.1, an update to its multi-model router, claiming performance gains up to 7.9 points over v1.0 and results that beat Anthropic's Fable 5 despite Fable 5 not being part of the router's model pool. All benchmark figures come from Sakana itself and remain unverified.