Inception: Mercury 2

Inception: Mercury 2, released on March 4, 2026, is positioned as an extremely fast reasoning LLM, leveraging a novel reasoning diffusion LLM (dLLM) architecture. It consistently performs among the fastest models, ranking in the 95th percentile across eight benchmarks, and offers competitive pricing, placing in the 51st percentile. Notably, Mercury 2 demonstrates exceptional reliability with a 100% success rate across all benchmarks, indicating consistent operational stability. The model exhibits significant strengths in speed-sensitive tasks and reasoning. It achieved 98.0% accuracy in Hallucinations (Baseline) and 96.0% in Reasoning (Baseline), ranking among the most accurate models in its speed class for both, and notably in the top 3 for speed in Reasoning. Its Email Classification accuracy was also strong at 97.0%, again being the most accurate among models of comparable speed. However, Mercury 2 shows notable weaknesses in General Knowledge (8.0% accuracy) and performs below average in Coding (67.0% accuracy) and Mathematics (58.0% accuracy). Instruction Following (53.6% accuracy) and Ethics (94.0% accuracy) also present areas for improvement relative to its overall speed prowess.