GPT-5-Codex extends agentic coding runs beyond seven hours
OpenAI released GPT-5-Codex across Codex cloud, terminal, IDE, GitHub, and mobile surfaces. The company reported successful independent software-engineering runs lasting more than seven hours, while its system-card addendum documented default sandboxing, disabled network access, prompt-injection training, and High biological and chemical capability treatment.
0 comments · 0 votes
Sign in to join the discussion →
No comments yet. Start the discussion.
Why it moved the index
A released coding agent that can plan, edit, test, and recover over multi-hour tasks materially lengthens practical machine autonomy. Broad product availability and the High biological and chemical capability classification strengthen the nexus. The score is moderated because the longest-run result is developer-reported and default sandbox, network, and review controls constrain external effects.
Assessment history
-
R1
Toward 62 · confidence 91
New dated primary-source model launch and system-card evidence absent from the durable catalogue.
14 Aug 2026
Share this page
-
DoomBench assesses “GPT-5-Codex extends agentic coding runs beyond seven hours” as evidence moving toward doom, with magnitude 62 and confidence 91 out of 100 in the autonomy and agency category.
-
The DoomBench assessment of “GPT-5-Codex extends agentic coding runs beyond seven hours” is based on reporting from OpenAI and records the editorial rationale, source quality, attribution, and revision history.
-
DoomBench summarizes “GPT-5-Codex extends agentic coding runs beyond seven hours” as follows: OpenAI released GPT-5-Codex across Codex cloud, terminal, IDE, GitHub, and mobile surfaces. The company reported successful independent...
https://www.doombench.com/news/gpt-5-codex-extends-agentic-coding-runs-beyond-seven-hours-2025-09-15