Taste-Bench benchmark reveals weak long-term decision-making in AI agents; top model scores only 59.7%
New Taste-Bench benchmark measures agent long-term decision preferences across coding and AI research tasks using 502 branching scenarios, testing 14 models from the Claude, GPT, Grok, DeepSeek, GLM, MiniMax, and Mistral families.
Return to original text
A short introduction is retained here; the full text from the source website is not reposted. For specific wording and subsequent updates, please refer to the original article.
Open AIHOT original