Inception: Mercury Coder

Inception: Mercury Coder, launched on April 30, 2025, is a groundbreaking diffusion large language model (dLLM) that consistently performs among the fastest models, ranking in the 99th percentile for speed across various benchmarks. It also offers competitive pricing, typically providing cost-effective solutions (62nd percentile), and demonstrates exceptional reliability with minimal technical failures (98th percentile). Across benchmark categories, Mercury Coder showcases remarkable speed, securing top 3 positions in Coding, Email Classification, and General Knowledge, and achieving #1 in Reasoning and Ethics, where it is noted as a "Speed champion" with near-perfect accuracy. While its speed is a clear strength, its accuracy varies. It achieved 80.0% in Coding (50th percentile) and 58.0% in Reasoning (49th percentile), indicating average performance in these areas. Its 92.0% accuracy in Email Classification, however, places it at the 21st percentile, suggesting a relative weakness in this specific classification task despite its speed. Conversely, it excels in Ethics (98.0% accuracy) and General Knowledge (93.5% accuracy), demonstrating strong capabilities in these domains. Its breakthrough discrete diffusion approach enables 5-10x faster operation than speed-optimized competitors, making it highly appealing for rapid iteration and responsive code completion.