GPT 5.3 Codex now official

OpenAI has unveiled GPT 5.3 Codex, described as its most capable agentic coding model to date. This was released just minutes after Anthropic’s Opus 4.6 announcement, tuned for autonomous coding workflows.

Among its uses include debugging, running tests, managing deployments, and operating within full OS environments. OpenAI says early versions of 5.3 Codex were used to help build the final model itself, from debugging its own training runs to managing evaluations.

The coding model itself is an early form of recursive self‑improvement in production tooling. Benchmarks like SWE-Pro and TerminalBench show large gains over GPT 5.2 Codex, with 5.3 achieving a higher SWE-Pro score than Opus 4.5 and beating Opus 4.6 on TerminalBench by a wide margin.

In demos, 5.3 Codex builds full 3D racing games with multiplayer and drifting, underwater exploration titles, and complete PowerPoint decks, Word documents, and spreadsheets from natural language prompts. It also posts strong results on OSWorld, which measures a model’s ability to control real desktop interfaces.

Right now, GPT 5.3 Codex is only available through paid ChatGPT plans, with API access “coming soon.” For devs in PH and SEA, it’s poised to be the default pick for heavy agentic software automation once API access opens.