GPT-5.2-Codex is OpenAI’s most advanced agentic coding model yet, delivering major gains in long-horizon software engineering and defensive cybersecurity while introducing new safeguards to manage its growing real-world impact. Sam Altman is depicted above. (Source: Image by RR)

Benchmark Results Show Strong Gains in Realistic Terminal Environments

OpenAI has unveiled GPT-5.2-Codex, its most advanced agentic coding model to date, designed for professional software engineering and defensive cybersecurity. Built as a specialized version of GPT-5.2, the new model introduces major improvements in long-horizon coding tasks, including context compaction, large-scale refactors, migrations, and better performance in Windows environments. The release marks a significant step forward in AI-assisted development, positioning Codex as a more reliable collaborator on complex, real-world engineering projects.

The model, as noted in openai.com, achieves state-of-the-art results on benchmarks such as SWE-Bench Pro and Terminal-Bench 2.0, demonstrating stronger performance in realistic repository and terminal environments. GPT-5.2-Codex is better at maintaining context across extended sessions, adapting to changing plans, and completing multi-step workflows without losing coherence. Enhanced vision capabilities also allow the model to interpret screenshots, UI mockups, and technical diagrams, accelerating the path from design concepts to production-ready code.

Beyond software engineering, GPT-5.2-Codex represents a notable leap in cybersecurity capability. OpenAI reports a third major jump in performance across professional Capture-the-Flag evaluations, following earlier gains in GPT-5-Codex and GPT-5.1-Codex-Max. While the model does not yet reach OpenAI’s “High” cyber capability threshold, it has already proven effective in real-world defensive research, including assisting security engineers in discovering previously unknown vulnerabilities in widely used frameworks like React.

Recognizing the dual-use risks of increasingly capable AI systems, OpenAI is pairing the rollout with additional safeguards, tighter access controls, and a new trusted access pilot for vetted security professionals. GPT-5.2-Codex is now available across Codex surfaces for paid ChatGPT users, with API access planned in the coming weeks. OpenAI says lessons from this deployment will guide future releases as agentic AI systems continue to advance toward more powerful — and potentially sensitive — cyber capabilities.

read more at openai.com