OpenAI drops GPT-5.3-Codex for autonomous technical agency

OpenAI is pivoting from code assistance to full-spectrum technical agency, marking the first time a model has been instrumental in its own creation and deployment.

• –Reaches 57% on SWE-bench Pro and 64.7% on OSWorld-Verified, nearly doubling previous benchmarks for real-world computer operation.
• –The new Codex-Spark variant enables real-time coding and instant infrastructure scaling via the Frontier operating system.
• –Classified as "High capability" for cybersecurity, signaling a major leap in reliability for autonomous vulnerability research and patching.
• –Direct competition with Anthropic’s Claude 4.6 in the "Coding Agent War" for enterprise engineering dominance.