The competition at the top of the AI industry has intensified this week as Anthropic released Claude Opus 4.6, only to be instantly challenged by OpenAI’s new model – GPT-5.3 Codex. The new launch from OpenAI, GPT-5.3-Codex, is an AI model designed to take on far more complex and long-running tasks than earlier versions.
OpenAI said the model represents a major step forward in how AI systems can assist developers and professionals, not just by writing code, but by reasoning through problems, using tools and working alongside humans in real time. Anthropic too has claimed similar improvements, setting up one of the most direct model-to-model rivalries yet.
According to OpenAI, GPT-5.3-Codex combines the strongest coding abilities of its previous Codex model with the broader reasoning and professional knowledge found in its latest GPT-5 series. The result is a system that can handle large projects that stretch over hours or even days while remaining responsive to user feedback throughout the process. The company also claims the new model runs around 25 per cent faster than its predecessor.
Unlike earlier tools that required users to wait for a finished result, GPT-5.3-Codex is designed to behave more like a colleague. Users can check in while it is working, ask questions, change direction or request improvements without losing the context of the task.
Strong gains in coding and real-world tasks
OpenAI says GPT-5.3-Codex now leads industry benchmarks that measure real-world software engineering and computer use. It performed strongly on tests that assess how well an AI can operate a terminal, manage files, navigate operating systems and complete practical tasks across multiple programming languages.
Beyond traditional coding, the model has also shown improvements in web development. In internal tests, GPT-5.3-Codex was able to build full web applications and games from scratch, improving them over time with minimal prompting. Compared with earlier versions, it produces more polished layouts, sensible defaults and production-ready features, even when given vague instructions.
On the other hand, Anthropic’s Opus 4.6 has achieved the highest score so far on the widely followed Terminal Bench 2.0 benchmark, which is a test designed to assess real-world coding and command-line problem-solving. However, it now seems Anthropic’s lead was short-lived, as within hours OpenAI also claimed its new model is about 25 per cent faster than GPT-5.2 Codex while at the same time combining stronger reasoning abilities.
Story continues below this ad
More than just a coding assistant
OpenAI has described GPT-5.3 Codex as an all-rounder capable of handling extended research, tool use, and complex execution tasks, making it ideal for long-running coding workflows. The AI powerhouse says the new model is built to support the full software lifecycle. That includes writing documentation, analysing data, preparing presentations, drafting reports and assisting with research.
In evaluations that measure professional knowledge work across dozens of occupations, GPT-5.3-Codex matched the performance of OpenAI’s top general models. The company believes this broad capability makes Codex useful not just for developers but also for designers, product managers, analysts and researchers who rely on computers for daily work.
Security, availability and what comes next
Given its growing power, OpenAI has placed added emphasis on cybersecurity safeguards. GPT-5.3-Codex is the first model the company classifies as high-capability for cybersecurity tasks, and it has been trained to help identify software vulnerabilities. OpenAI says additional monitoring and access controls are in place to reduce misuse while supporting defensive security research.
GPT-5.3-Codex is now available to users on paid ChatGPT plans through the Codex app, command-line tools and IDE extensions. API access is expected to follow later.
Story continues below this ad
With this release, OpenAI signals a broader ambition: moving from AI that simply writes code to systems that can plan, execute and collaborate across an entire computer workflow. If successful, GPT-5.3-Codex could mark a turning point in how people work with AI, not as a tool that responds, but as one that actively participates.
Regardless of the successive announcements by OpenAI and Anthropic, it is still unclear who is the clear winner here. This is because independent benchmark comparisons are limited and the two companies are using different testing methodologies in their public reports.
