Model Comparison: GPT‑5.3 Codex vs GPT‑5.2 vs GPT‑5.2 Codex

Editable comparison sheet. Click any cell to edit. Use Highlight differences to quickly spot what changed. Data is stored locally in your browser.

Highlight differences

Comparison table

I don’t have authoritative public specs for these exact model names. This page is designed so you can paste your internal benchmarks/pricing/context limits and keep a tidy, consistent comparison.

Metric	GPT‑5.3 Codex	GPT‑5.2	GPT‑5.2 Codex

Quick summary

Import / Export (JSON)

Export your table to JSON (for sharing or versioning) and import it back here. You can also paste JSON into a URL query string via the button in the header.

How to use

Researched facts (public)

GPT‑5.3‑Codex (Feb 5, 2026): Official system card says it is the first to pass all thresholds across three cybersecurity evaluations, and it matches GPT‑5.2‑Codex on a professional CTF set while benefiting from compaction for long-horizon work.
GPT‑5.2‑Codex (Dec 18, 2025): Official addendum describes product mitigations like an agent sandbox and default-disabled network access in the hosted container, and attributes strong performance on professional CTFs in part to compaction.
GPT‑5.2 (Dec 11, 2025): Secondary source (Simon Willison) reports a 400,000 token context window and 128,000 max output tokens, citing OpenAI docs.

Links: GPT‑5.3‑Codex system card (PDF) · GPT‑5.2‑Codex addendum (PDF) · Simon Willison on GPT‑5.2 · Simon Willison on GPT‑5.3‑Codex release

Fill in the cells with your real measurements (or qualitative ratings).
Toggle “Highlight differences” to spot where models diverge.
Export JSON to share with teammates or paste into docs.