Particle.news
Download on the App Store
17 ARTICLES
·
3h ago

OpenAI Releases GPT-5.3-Codex, a Faster Agentic Coding Model That Helped Build Itself

OpenAI is gating sensitive automation under a new cybersecurity program due to the model’s elevated risk classification.

Image
Credit: The New Stack.
Image
SUN VALLEY, IDAHO - JULY 08: CEO of OpenAI Sam Altman arrives at the Sun Valley lodge for the Allen & Company Sun Valley Conference on July 8, 2025 in Sun Valley, Idaho. Every year, some of the world's wealthiest and most powerful figures from the media, finance, technology, and political spheres converge at the Sun Valley Resort for the exclusive week-long conference hosted by boutique investment bank Allen & Co.

Overview

  • GPT-5.3-Codex is available now to paid ChatGPT users across the Codex app, CLI, IDE extension, and web, with API access to follow under restrictions for high-risk cybersecurity uses.
  • OpenAI says the model runs about 25% faster than GPT-5.2-Codex and is built for long-running, steerable workflows that can handle full software-lifecycle tasks and operate for days without losing context.
  • The company reports that early versions of GPT-5.3-Codex were used to debug training, manage deployment, and analyze evaluations, accelerating the model’s own development.
  • OpenAI cites strong internal results on TerminalBench 2.0 (77.3%), SWE‑Bench Pro, and OSWorld‑Verified, though independent, cross-lab benchmarking is still being sought.
  • Classified by OpenAI as “high-capability” for cybersecurity, the release includes a Trusted Access for Cyber pilot, expanded monitoring and safety training, and $10 million in API credits to support defensive research, as Anthropic’s Opus 4.6 launched minutes earlier in a tightening rivalry.