OpenAI made GPT-5-Codex the default model across Codex CLI, the Codex IDE extension, and Codex cloud environments as of April 4. If you’ve been using Codex recently and noticed a behavioral change, this is why.

GPT-5-Codex is not the same model as GPT-5. It’s a variant of GPT-5 specifically fine-tuned for agentic coding tasks — designed to perform well on the kind of multi-step, tool-using work that defines how developers actually use AI coding assistants today.

What GPT-5-Codex Is (And Isn’t)

The naming is confusing by design: OpenAI is reusing the Codex brand — originally the model that powered GitHub Copilot in 2021 — for a modern, GPT-5-based successor. The new GPT-5-Codex shares little with its predecessor architecturally. It’s a GPT-5 derivative with training specifically targeting:

  • Long-horizon task completion across multiple files and directories
  • Accurate tool use in agentic contexts (shell commands, file operations, API calls)
  • Code generation, debugging, and refactoring with explicit reasoning
  • GitHub repository navigation and multi-file context understanding

It’s available exclusively through the Responses API — not the Chat Completions API. This is deliberate: the Responses API supports streaming, tool calls, and structured outputs natively, which are the primitives that agentic coding workflows actually need. Developers building on top of GPT-5-Codex via the API need to update their integration accordingly.

Where It’s Available

Codex CLI: The command-line interface for running Codex-powered coding agents in a terminal. GPT-5-Codex is now the default; the previous default (a GPT-4o variant) remains available as a selectable option.

Codex IDE extension: Available for VSCode and JetBrains. Auto-updated to GPT-5-Codex as the default agent model for agentic tasks (standard autocomplete continues to use a faster, lighter model).

Codex cloud environments: OpenAI’s managed cloud execution environments for running Codex agents on longer tasks. GPT-5-Codex brings improved performance on the benchmark tasks OpenAI uses to evaluate these environments.

GitHub integration: GPT-5-Codex is available as the model for GitHub Copilot’s agentic workspace features, though rollout there is phased.

The Competitive Context

Claude Code and GPT-5-Codex are now the two dominant commercial agentic coding products. They take different architectural approaches: Claude Code runs locally with optional cloud subagents (including the new Ultraplan feature), while Codex is primarily a cloud-managed service with local CLI access. Both are converging on similar use cases — long-horizon coding tasks that require planning, multi-file context, and tool use.

The Claude Code vs. GPT-5-Codex comparison is one developers are already running in community benchmarks. Early results are mixed, with neither model dominating across all task types. For Claude Code users, the most relevant data point: GPT-5-Codex is now a serious competitor rather than a legacy product getting minor updates.

GPT-5-Codex is available via the OpenAI API at developers.openai.com/api/docs/models/gpt-5-codex. Pricing is aligned with GPT-5 API pricing for the equivalent tier.

Sources: OpenAI — GPT-5-Codex model page · openai.com/index/introducing-upgrades-to-codex