On 16 May 2025 OpenAI released a research preview of Codex, a cloud agent powered by codex-1 (an o3 variant trained for software engineering) that writes features, runs tests, and opens pull requests in a sandboxed repo.
Current score
+1.11
10 base · Major (tier 3 of 5, 10 pts)
× 0.7000 attribution · Critical contribution
× 0.3500 evidence · Technical disclosure or preprint
× 0.7000 realization · Independently validated or deployed
× 0.6500 durability
Event-level product before credit split: 1.11
A cloud coding agent that opens PRs is a major product change (tier 3). Credit goes to the named 2025 codex-1 release, not to GPT-6 Astra or 2021 Codex (0.70). Evidence is a lab announcement plus press (0.35). Realization is a paid-tier research preview, not universal GA (0.70).
What happened
This is a different named system from 2021 Codex. The 2025 product is an asynchronous agent: users assign tasks from ChatGPT, each job runs in an isolated cloud environment with the repository loaded, and the agent can edit files, run linters and tests, and propose PRs. OpenAI rolled it out first to Pro, Business, and Enterprise, with CLI and GitHub integration. TechCrunch and OpenAI both describe it as part of the 2025 agentic-coding wave alongside Claude Code and GitHub Copilot’s coding agent. There is no independent measured productivity study yet, so the score is for a significant product launch that changed how frontier labs ship coding tools, not for a proven output multiplier.