Taste-Bench: AI Agents Flunk Long-Horizon Strategy Decisions

Taste-Bench shows GPT-5.6 Sol picks the right strategic path on under 60% of binary choices — and longer thinking doesn't improve the score.