The recent scores for the Image-to-WebDev competition in Code Arena have revealed that Opus 5 (Max) leads the rankings with 1,669 points, followed by GPT-5.6 Sol and Grok-4.5, which scored 1,581 and 1,578 points, respectively. This competition evaluates AI models on their ability to generate functional websites from images and screenshots, utilizing multi-step reasoning and tool use as part of the agentic coding workflows. The results showcase the effectiveness of these models in the iterative development process essential for website creation.

Aryan: Aryan is a founding engineer associated with the Code Arena platform. He provides detailed explanations of the Image-to-WebDev benchmark, including its focus on AI models’ abilities in visual web development and agentic workflows.
Opus 5: Opus 5 (Max) is an advanced AI model focused on high-performance reasoning and coding capabilities. It is evaluated in the Image-to-WebDev benchmark within Code Arena for its ability to generate websites from images and screenshots through agentic workflows. The model is positioned at the top of the latest rankings in this specific challenge.
Kimi K3: Kimi K3 (Max) is an advanced large language model variant optimized for versatile AI applications including development tasks. It is included in the Code Arena Image-to-WebDev rankings, highlighting its capabilities in image-based web generation and iterative agentic workflows.
Grok-4.5: Grok-4.5 is an advanced AI model from xAI designed for strong performance in coding and problem-solving domains. It participates in the ongoing Image-to-WebDev benchmark in Code Arena, where models are assessed on website generation from images alongside agentic coding processes.
GPT-5.6 Sol: GPT-5.6 Sol is a high-performance configuration of OpenAI’s GPT-5.6 model tailored for complex tasks. It competes in the Code Arena’s Image-to-WebDev evaluation, demonstrating strengths in converting visual inputs into functional web development outputs via multi-step reasoning and tool integration.
GPT-5.6 Luna: GPT-5.6 Luna is a distinct variant of OpenAI’s GPT-5.6 model used in benchmarking exercises. It takes part in the Code Arena’s Image-to-WebDev assessment, focusing on website creation from images and screenshots through advanced reasoning and tool-assisted processes.
GPT-5.6 Terra: GPT-5.6 Terra is a configuration of OpenAI’s GPT-5.6 model suited for specialized evaluation scenarios. It is ranked in the Code Arena Image-to-WebDev challenge for its ability to handle image-to-website generation and associated agentic coding tasks.
Muse Spark 1.1: Muse Spark 1.1 is an AI model variant developed for creative and technical coding challenges. It appears in the latest Code Arena Image-to-WebDev leaderboard, where its performance is measured against other models in generating websites from visual prompts using multi-step reasoning.

Benchmarking: Code Arena evaluates multiple AI models on their capacity to convert images and screenshots into functional websites.
Agentic Coding: The Image-to-WebDev task assesses models using multi-step reasoning combined with tool use in iterative development processes.