Codex 5.3

Codex 5.3 is a coding-specialised AI model developed by OpenAI, positioned as a coding and software development assistant. In the UK AI Security Institute's independent evaluation of Claude Mythos Preview on 15 April 2026, Codex 5.3 was used as one of three comparison models — alongside Claude Opus 4.6 and GPT-5.4 — to benchmark Mythos on isolated capture-the-flag (CTF) cybersecurity tasks. Mythos scored above 85% on those tasks; Codex 5.3 fell within 5 to 10 percentage points, establishing it as competitive on single-task discrete security benchmarks.

Codex models have historically been the primary benchmark target for coding-task comparisons in the AI safety and capability evaluation community, making Codex 5.3's inclusion in the AISI CTF battery a standard methodology choice. The AISI evaluation did not assess Codex 5.3 on the 32-step autonomous attack chain benchmark where Mythos demonstrated a genuine capability advantage — that benchmark was Mythos-specific.

The significance for this beat is context rather than substance: Codex 5.3's performance within 5 to 10 points of Mythos on single tasks confirms that the public frontier is close to the restricted frontier on discrete benchmarks. The gap that matters — multi-step autonomous execution — is not captured in the CTF scores published by AISI.