Step Intelligence has unveiled Step Image Edit 2, a new image generation and editing model featuring 3.5 billion parameters. Despite its smaller size, the model outperforms open-source models with 12 to 20 billion parameters in real-world tests. It supports text-to-image generation and image editing, offering features like Chinese and English rendering, local editing, and style transfer. The model ranks first on the KRIS-Bench benchmark for lightweight models. Step Image Edit 2 introduces innovative techniques such as Multi-Expert Self-Evolution Learning and Distribution-Matching Reinforcement Learning, enhancing performance without increasing parameters. The model processes images in 0.5 to 2 seconds, with API pricing set at CNY 0.02 per image, and is free for the first week post-launch. Over 50 million specialized training samples were used, including 20 million for text editing scenarios.