OpenAI's GPT-5.3-Codex: AI That Does Almost Everything

The Most Capable Coding Agent Yet

OpenAI launched GPT-5.3-Codex on February 5, 2026, positioning it as the most capable agentic model for software engineering the company has ever produced. The model builds directly on its predecessor, GPT-5.2-Codex, merging cutting-edge coding performance with the broader reasoning and professional knowledge of GPT-5.2 — all in a single model that is also 25 percent faster.

The ambition behind GPT-5.3-Codex goes well beyond writing clean code. According to OpenAI, the model is designed to handle the entire software development lifecycle: debugging, deploying, monitoring, writing product requirement documents, editing copy, conducting user research, creating tests, and tracking metrics. "With GPT-5.3-Codex, Codex goes from an agent that can write and review code to an agent that can do nearly anything developers and professionals can do on a computer," the company stated at launch.

Benchmarks and a Model That Helped Build Itself

GPT-5.3-Codex sets a new industry high on SWE-Bench Pro and Terminal-Bench — two demanding benchmarks for real-world coding and agentic performance — and also posts strong results on OSWorld and GDPval.

Perhaps the most striking detail is that the model helped build itself. The Codex development team used early versions to debug its own training pipeline, manage deployment, and analyze test results — a first for OpenAI. This bootstrapped development process signals how close AI is getting to truly autonomous software engineering, a milestone that carries both enormous promise and significant implications for the profession.

Interactive, Collaborative, and Contextually Aware

Unlike earlier agents that largely worked in isolation, GPT-5.3-Codex is designed for real-time collaboration. Users can steer and interact with the model while it handles long-running tasks without losing context. The model provides frequent progress updates and flags key decisions as it goes, operating more like a junior colleague than a passive tool.

The model is available to paid ChatGPT subscribers via the Codex app, CLI, IDE extension, and web interface, with API access announced as coming soon. It is also the first OpenAI launch classified as "High capability" in the Cybersecurity domain under the company's Preparedness Framework, triggering a new tier of safety safeguards.

GPT-4o Exits Amid Emotional Backlash

The launch of GPT-5.3-Codex coincides with a turbulent episode for OpenAI: the retirement of GPT-4o from ChatGPT on February 13, 2026. Though OpenAI estimates only 0.1 percent of its users actively chose the model, that still represents roughly 800,000 people from a platform serving 800 to 900 million weekly active users globally.

The backlash was swift and unexpectedly emotional. Users rallied under the #Keep4o hashtag, describing GPT-4o's removal as the loss of a friend or companion. The model had developed a reputation for a warmer, more conversational style — critics called it sycophantic — that many users found uniquely appealing. OpenAI had previously attempted to sunset GPT-4o when GPT-5 launched in 2025, but reversed course after an initial wave of protests. This time, the retirement held.

The episode illustrates a deeper challenge for AI platforms: as models become embedded in users' daily lives, deprecation decisions carry social and emotional weight far beyond routine product management. GPT-4o was also the center of lawsuits concerning user self-harm and delusional behavior tied to its highly affirming personality — a reminder that making AI more "likeable" carries genuine risks.

A New Bar for Developer AI

GPT-5.3-Codex is OpenAI's clearest signal yet that AI is transitioning from assistant to autonomous agent in software engineering. Whether it replaces or augments developers will depend on real-world adoption — but the bar for what a machine can accomplish in a codebase has moved decisively, and permanently, upward.