OpenAI launches GPT-5.3-Codex, 25% faster and broader in scope

Zdnet reports that OpenAI announced GPT-5.3-Codex, which the company describes as its most capable agentic coding model yet, running 25% faster and capable of debugging and deploying parts of itself during development. OpenAI says the model now supports more than just coding, covering work across the software lifecycle including debugging, deploying, monitoring, writing PRDs, editing copy, user research, tests and metrics.

The company highlights that Codex can be steered mid-task without losing context, handle longer-running processes, and produce richer outputs from underspecified prompts. The company reports GPT-5.3-Codex sets new industry highs on benchmarks such as SWE-Bench Pro and Terminal Bench, shows strong performance on OSWorld and GDPVal, and uses fewer tokens for assignments.