Mistral: Devstral Medium

Mistral's Devstral Medium is a high-performance code generation and agentic reasoning model, demonstrating strong capabilities across various benchmarks. It performs among the fastest models, ranking in the 69th percentile for speed, and offers competitive pricing, placing in the 51st percentile. Notably, Devstral Medium exhibits exceptional reliability with a 100% success rate across all benchmarks, indicating minimal technical failures. The model excels in several areas, achieving perfect accuracy in Hallucinations, General Knowledge, and Ethics, often being the most accurate at its price point and speed. It also performs perfectly in Email Classification. Its coding prowess is evident with an 88.0% accuracy in the Coding benchmark, and a respectable 90.0% in Mathematics, where it is also noted as the most accurate among models of similar speed. While strong in Instruction Following (65.0% accuracy) and Reasoning (64.0% accuracy), these areas represent relative weaknesses compared to its perfect scores in other categories. Its primary strength lies in its robust code-related performance (61.6% on SWE-Bench Verified), positioning it ahead of competitors like Gemini 2.5 Pro and GPT-4.1.