Coding-specialized variant of GPT-5.3, tuned for agentic IDE workflows.
OpenAI's specialized self-improving coding model with state-of-the-art software engineering performance
| Benchmark | GPT-5.3 Codex | GPT-5.2 Codex | Δ |
|---|---|---|---|
| Terminal-Bench | 77.3% | 72.8% | +4.5 |
| SWE-Bench Pro | SOTA | — | — |
| Speed | 1,000+ tok/s | — | — |
| SWE-bench Verified | 82.4% | 78.2% | +4.2 |
| HumanEval | 96.8% | 95.1% | +1.7 |
| LiveCodeBench | 84.2% | 80.4% | +3.8 |
| Aider Polyglot | 79.5% | — | — |