OpenAI logo./Courtesy of OpenAI

OpenAI said on the 6th that it unveiled a new model, GPT-5.3-Codex, which significantly strengthens agent-style coding capabilities.

The newly unveiled GPT-5.3-Codex is an artificial intelligence (AI)-based coding model that combines the frontier-level coding performance of GPT-5.2-Codex with the reasoning and domain knowledge capabilities of GPT-5.2. Processing speed improved by about 25% compared with the previous version.

According to OpenAI, GPT-5.3-Codex posted industry-leading performance on SWE-Bench Pro and Terminal Bench, and showed strong results on major benchmarks such as OSWorld and GDPVal. These benchmarks comprehensively evaluate coding ability, agent capabilities, and real-world task performance.

GPT-5.3-Codex was also used in the model's own development process. The OpenAI Codex team used early versions to debug internal training, manage deployments, and diagnose test results, improving the model's completeness.

GPT-5.3-Codex is available to paid ChatGPT users across all Codex-supported environments (app, CLI, IDE extension, web). OpenAI said it will sequentially provide safe API access in the future.

Following the recent release of the Codex app, improved to manage multiple agents simultaneously, OpenAI further expanded Codex's use cases by unveiling this model. As a result, Codex is evolving beyond simple code writing and reviews into an agent that autonomously handles a wide range of computer-based tasks performed by developers and experts.

This update also ties into partnership-based development environments such as Apple Xcode 26.3 and GitHub. Developers can use Codex directly within Xcode and GitHub workflows to carry out complex tasks more efficiently.

According to OpenAI, downloads of the Codex app exceeded 500,000 as of on the 2nd, and weekly active users are at 1 million. For a limited time, Codex is available on the free and Go plans of ChatGPT, and usage limits for the Plus, Pro, Business, Enterprise, and Edu plans have doubled compared with the previous limits. These limits apply equally across all environments, including app, CLI, IDE, and cloud.

※ This article has been translated by AI. Share your feedback here.