Aurora Alpha

Aurora Alpha, provided by openrouter, is a reasoning model designed for speed, created on February 8, 2026, with a context length of 128,000 tokens. It exhibits moderate speed performance, ranking in the 36th percentile across eight benchmarks. Price data is currently unavailable, suggesting potential free tier usage. A significant strength is its exceptional reliability, boasting a 97% success rate across all benchmarks, indicating consistent and usable responses. The model demonstrates strong performance in Instruction Following (88th percentile) and Reasoning (80th percentile), even achieving a top 3 ranking in cost for reasoning tasks. It also shows solid accuracy in Email Classification (97.1%) and a good ability to acknowledge uncertainty in Hallucinations (90.0%). However, Aurora Alpha struggles significantly with General Knowledge (11th percentile) and Coding (20th percentile), suggesting these are notable weaknesses. Its performance in Ethics (26th percentile) and Mathematics (32nd percentile) is also below average. While designed for speed, its benchmark durations vary, with some tasks being relatively fast and others slower.