OpenAI has confirmed the general availability of GPT-5-Codex on the Responses API . The model can also be directly accessed through the Codex CLI using an API key . Official descriptions describe it as a specialized version of GPT-5 for agentic coding, suitable for scenarios such as long-term code review, refactoring, and multi-tool collaboration.
GPT-5-Codex maintains the same pricing as GPT-5. The model is released as "only available on the Responses API," with a statement stating that the underlying model snapshots will be regularly updated. Codex also includes updates to the default model and usage visibility, along with new instructions and an automatic compression mechanism to improve long-session stability.
Frequently Asked Questions
Q: On which interfaces is it available?
A: Responses API; Codex CLI can also use gpt-5-codex via API Key.
Q: What are the differences with GPT-5?
A: The same generation base optimizes reasoning and tool calls for encoding and proxy scenarios.
Q: What is the price?
A: On par with GPT-5. Please refer to the model page and invoice for details.
Q: Why only in Responses API?
A: Responses provides event streaming and managed tools, which are more suitable for "reason-to-action" workflows.
Q: How do I switch to the CLI?
A: Use /model gpt-5-codex or the startup parameter --model gpt-5-codex in Codex CLI.
