Particle.news
Download on the App Store

OpenAI Releases GPT-5.3-Codex, a Faster Agentic Coding Model With Guardrails

OpenAI is gating high‑risk uses under a Trusted Access pilot due to a high‑capability cybersecurity classification.

Overview

  • GPT-5.3-Codex is available now to paid ChatGPT users across the Codex app, CLI, IDE extensions and the web, with broader API access planned.
  • OpenAI says early versions of the model helped debug training runs, manage deployment and assess evaluations, accelerating its own development.
  • The company reports roughly 25% faster throughput versus GPT-5.2-Codex and leading internal results on SWE‑Bench Pro, Terminal‑Bench 2.0 and OSWorld‑Verified.
  • The release targets end‑to‑end workflows: the model can be steered mid‑task, run for extended periods and support work beyond coding such as docs, research and presentations.
  • Citing elevated cyber risk, OpenAI classifies the model as high capability for cybersecurity, is restricting advanced uses, and is offering $10 million in API credits to support defensive research as competition with Anthropic intensifies.