OpenAI's GPT-5 Codex Works Independently for Over 7 Hours on Complex Coding Tasks

OpenAI has recently introduced GPT-5 Codex, a specialized version of its flagship GPT-5 model, optimized for advanced software engineering and agentic coding. The new model, released in mid-September 2025, has been highlighted by industry observers for its ability to autonomously handle complex programming challenges and sustain long-duration tasks without frequent interruptions. This development marks a significant step in AI-assisted software development, aiming to provide a more capable and integrated coding partner.

One user, Haider., recently shared on social media, "> GPT-5 Codex is slow but a game-changer," further noting, "> it excels at hard C++ problems and algorithmic analysis, and it follows a task much longer before pausing." This observation aligns with OpenAI's design goals, as GPT-5 Codex was trained specifically on real-world engineering tasks, including building projects from scratch, debugging, and large-scale refactoring. The model demonstrates adaptive reasoning, dynamically adjusting its processing time to task complexity.

In contrast to its general-purpose counterpart, GPT-5 Codex is engineered for streamlined execution in coding environments. Haider. pointed out a key distinction, stating, "> unlike GPT-5, which sometimes stops for confirmation about 3x as often, which is super annoying." While GPT-5 utilizes a "router" to direct queries to various sub-models, potentially leading to more frequent confirmation steps, GPT-5 Codex is optimized to work independently for extended periods, with testing showing it can operate for over seven hours on complex tasks. For simpler tasks, it uses 93.7% fewer tokens than GPT-5, while for the most complex, it spends twice as long reasoning.

The new model is deeply integrated into OpenAI's Codex ecosystem, supporting developers through the Codex CLI, IDE extensions for platforms like VS Code and Cursor, GitHub, and the ChatGPT iOS app. Its capabilities extend to superior code reviews, generating 70% fewer incorrect comments than GPT-5, and it is described as "everything at once... Agent, LLM" by Haider., reflecting its autonomous, agentic approach to software development. This positions GPT-5 Codex as an evolving AI agent capable of planning and executing multi-step programming tasks.

GPT-5 Codex is available to users with ChatGPT Plus, Pro, Business, Edu, and Enterprise plans, with API access for developers rolling out soon. Its release follows the general availability of GPT-5 on August 7, 2025, which itself brought advancements in reasoning, multimodal capabilities, and a 256,000-token context window. OpenAI's dual release strategy underscores its commitment to both general-purpose AI and highly specialized, agentic models for specific domains like software engineering.