release

Addendum to GPT-5 system card: GPT-5-Codex

For builders automating coding workflows, GPT-5-Codex offers a model that can match its reasoning time to the problem at hand, making AI coding assistants more efficient and cost-effective across a range of tasks.

OpenAI Blog·September 14, 2025·1 min readrelease

releaseAddendum to GPT-5 system card: GPT-5-Codex

openai.com

What happened

OpenAI has published an addendum to its GPT-5 system card detailing a new model variant called GPT-5-Codex. According to the blog, GPT-5-Codex is a version of GPT-5 specifically optimized for agentic coding within the Codex environment. The key enhancement is dynamic adjustment of thinking effort: the model responds quickly to simple conversational queries or small tasks, and independently spends more time on complex tasks. This adaptive behavior aims to improve efficiency and autonomy in coding agents. For developers building AI workflows, this means a single model can handle both rapid-fire interactions and deep, multi-step programming challenges without manual tuning. The move reflects a broader trend in AI development toward models that can self-regulate computational resources based on task complexity, potentially reducing costs and improving user experience in automated coding assistants.

Key takeaways

OpenAI released GPT-5-Codex, a variant of GPT-5 optimized for agentic coding in the Codex environment.
The model dynamically adjusts its thinking effort based on task complexity: fast for simple queries, longer for complex tasks.
It is intended to improve efficiency by avoiding unnecessary compute on easy tasks while allocating more resources to difficult ones.
The model is described in an addendum to the GPT-5 system card, signaling a specialized release for coding agents.