Chat with GPT-5.3 Codex online

GPT-5.3 Codex is OpenAI's coding specialist, announced on 5 February 2026. It combines the coding strength of the Codex line with the general reasoning and professional knowledge of the wider GPT-5 series, and OpenAI's framing of it is deliberately broad: Codex moves from an agent that writes and reviews code to an agent that can do most of what a developer does on a computer.

That shows up in what it is tuned for. Debugging, deploying, monitoring, writing specs, editing copy, tests and metrics are all in scope, not just producing a function on request. It sets high marks on the benchmarks that measure real repository work rather than puzzle-solving — SWE-Bench Pro and Terminal-Bench in particular — and OpenAI reports it runs roughly 25% faster than GPT-5.2 Codex, which matters more than it sounds when a model is doing dozens of steps in a row.

It exposes adjustable reasoning effort, so the same model can answer quickly or think hard depending on what the task deserves, and it takes images as input, so a screenshot of a broken build or an unreadable chart can go straight into the thread. The context window runs to about 400K tokens with up to 128K tokens of output, enough to hold a large slice of a repository plus a long agent transcript.

On NinjaChat it is included in every plan. Pick it from the model list when the work is software, and switch to a general model when it is not.