Pricing, Performance & Features Comparison
Devstral Medium 2507 is a high-performance, code-centric large language model designed for agentic coding capabilities and enterprise use. It features a 128k token context window and achieves a 61.6% score on SWE-Bench Verified, outperforming several commercial models like Gemini 2.5 Pro and GPT-4.1. The model excels at code generation, multi-file editing, and powering software engineering agents with structured outputs and tool integration.
google/gemma-2-9b-it is a text-to-text, decoder-only large language model optimized for tasks such as question answering, summarization, and reasoning. It was trained on 8 trillion tokens with 9 billion parameters, aiming for efficient inference and strong performance despite its size. Its design supports flexible deployment for applications like chatbots, code generation, and research in NLP.