Sakana: Fugu Ultra vs. GLM 5.2 (runtimewire.com)

Sakana has announced the results of a competitive evaluation between its Fugu Ultra and GLM 5.2 models, with Fugu Ultra emerging victorious by a score of 107.9 to GLM 5.2’s 96.9. This decisive win indicates Fugu Ultra's superior performance, particularly in challenging tasks such as technical coding, debugging, and summarization, where it demonstrated greater fidelity to prompts and consistently robust outputs. While GLM 5.2 excelled in certain areas like localization and providing convincing justifications, these strengths did not compensate for its weaker performance across the broader range of tasks assessed, particularly in high-stakes scenarios. The evaluation methodology involved generating 12 fresh text tasks judged by OpenAI's GPT-5.6 Sol to ensure unbiased results. Fugu Ultra outperformed GLM 5.2 by effectively handling edge cases—such as cache bugs and data processing—while also achieving better summaries and precise proofreading. The testing signals a clear trend towards Fugu Ultra being the more dependable model for diverse real-world applications, underscoring its potential to support complex workflows in the AI/ML community.