OpenAI launches macOS app for agentic coding

Artificial intelligence is having a significant impact on software development, with much of the programming work now handled by agents. On Monday, OpenAI launched a new macOS application for Codex, its coding tool. The release represents an effort to incorporate agentic software development—defined as systems where AI agents work independently on coding tasks—into its product lineup. The launch follows the initial release of Codex as a command line tool last April, which was followed by a web interface expansion one month later. The new desktop application is designed to work with multiple agents in parallel, integrating agent skills and other workflows.

The new application integrates GPT-5.2-Codex, which is OpenAI’s most powerful coding model. The company is positioning the model to compete directly with established agentic coding applications like Claude Code and Cowork. To evaluate performance, developers look to coding benchmarks. On TerminalBench, a benchmark that tests how well AI handles command-line programming tasks, GPT-5.2 holds the top spot, at least as of press time. However, competing agents powered by Gemini 3 and Claude Opus have recorded roughly equivalent scores that remain within the margin of error of the benchmark. Similarly, results on SWE-bench, a coding benchmark that tests an AI’s ability to resolve real-world software bugs, show no clear advantage for the OpenAI model. This highlights a broader challenge in the industry, as agentic use cases remain difficult to benchmark effectively.

OpenAI CEO Sam Altman noted that while GPT-5.2 is the company’s strongest model for complex tasks, its previous interface made it difficult for developers to use. The new macOS app aims to resolve this by providing a more flexible interface. Altman explained that taking this level of model capability and putting it into a more flexible interface would matter quite a bit for developers. According to Altman, the interface allows users to build complex projects in a few hours. “You can use this from a clean sheet of paper, brand new, to make a really quite sophisticated piece of software in a few hours. As fast as I can type in new ideas, that is the limit of what can get built,” Altman said.

Why it matters

OpenAI is attempting to catch up in the agentic software development space by integrating its powerful GPT-5.2-Codex model into a more flexible macOS interface, aiming to compete with existing tools like Claude Code.