Back to glossary

Term

GPT-5.3-Codex

GPT-5.3-Codex is OpenAI's specialized agentic coding model (February 2026), the successor to GPT-5.2-Codex — 25% faster, with substantial gains on Terminal-Bench and OSWorld.

GPT-5.3-Codex — explained in more detail

OpenAI introduced GPT-5.3-Codex on February 5, 2026. It is a model in the GPT-5.3 line specialized for agentic coding and the successor to GPT-5.2-Codex. According to OpenAI it is the most capable Codex model to date: it does not just write and review code but performs nearly every action a developer would carry out on a computer — tool use, computer use, long workflow chains. Interactions are about 25% faster than with the predecessor.

Benchmarks and contrast with GPT-5.2-Codex

  • Terminal-Bench 2.0: 77.3% (vs. 64.0%) — the largest jump.
  • OSWorld-Verified: 64.7% (vs. 38.2%) — a new high for computer-using agents.
  • SWE-Bench Pro: 56.8% (vs. 56.4%) — modest improvement.
  • GDPval: 70.9%.
  • Cybersecurity CTF: 77.6% — the first OpenAI model rated “High capability” for cybersecurity tasks under their internal classification.

Availability

Available in paid ChatGPT tiers; API access is announced for once safety enablement is complete. A smaller variant optimized for real-time coding, GPT-5.3-Codex-Spark, followed on February 12, 2026 as a research preview.

Entdecke mehr

Themenuebersicht