GPT-5.3-Codex

GPT-5.3-Codex was released to the Responses API on 24 February 2026 and OpenAI still describes it as the most capable agentic coding model to date. It is built for Codex and similar environments rather than for general chat, and it is the only place in OpenAI's range where the hosted shell and skills tools are offered. The model takes text and images, returns text, and works with a 400,000-token context window split into a maximum of 272,000 input tokens and 128,000 output tokens. Reasoning effort has four levels - low, medium, high and xhigh - and reasoning tokens are supported. Two things set it apart from the rest of the catalogue. It is not available through Chat Completions at all: the Responses API is the only supported endpoint, and Batch, fine-tuning and the audio routes are closed to it. And its knowledge cutoff is 31 August 2025, roughly six months before the model itself shipped, so a coding model released in 2026 knows nothing about libraries or language releases from the autumn of 2025 onwards. Pricing is $1.75 per million input tokens, $0.175 cached and $14 per million output - identical to GPT-5.2-Codex, which OpenAI prints in a comparison table on the model's own page.

▸Videos

No videos yet.

▸Reviews