DeepSeek: R1 Distill Qwen 7B

DeepSeek: R1 Distill Qwen 7B, a 7 billion parameter model distilled from DeepSeek-R1, consistently ranks among the fastest models and offers highly competitive pricing across all evaluated benchmarks. This model, created on May 30, 2025, leverages reinforcement learning-enhanced reasoning data to transfer advanced capabilities into a smaller architecture. In terms of performance, the model demonstrates a notable strength in Instruction Following, achieving 34.3% accuracy, placing it in the 35th percentile. Its Coding performance is also respectable at 66.0% accuracy (27th percentile). However, a significant weakness is apparent in its performance across General Knowledge, Email Classification, and Ethics benchmarks, where it recorded 0.0% accuracy. This suggests a potential limitation in handling certain types of open-ended knowledge retrieval or nuanced classification tasks, or possibly an issue with the evaluation methodology for these specific categories. Despite these areas for improvement, its strong showing in mathematical benchmarks (92.8% pass@1 on MATH-500) and coding tasks (Codeforces rating 1189), as mentioned in its description, indicates a specialized proficiency in these domains. The model's efficiency in terms of speed and cost makes it an attractive option for applications where these factors are critical, particularly for tasks requiring strong instruction following and coding capabilities.