← Models

GPT-5-Codex

OpenAI · GPT-5

OpenAI's GPT-5 variant tuned for high-reliability coding, code review, and software engineering workflows.

Type
language
Context
400K tokens
Max Output
128K tokens
Status
current
Input
$1.25/1M tok
Output
$10/1M tok
API Access
Yes
License
proprietary
coding agentic software-engineering tool-use code-review refactoring
Released September 2025 · Updated February 15, 2026

Overview

Freshness note: Model capabilities, limits, and pricing can change quickly. This profile is a point-in-time snapshot last verified on February 15, 2026.

GPT-5-Codex is OpenAI’s coding-focused member of the GPT-5 family. It is optimized for software engineering workflows where correctness, repository awareness, and edit quality matter more than broad conversational style.

In model-routing terms, GPT-5-Codex is the specialist you call when coding quality is the top priority and you need consistent behavior over long code contexts.

Capabilities

Key strengths include:

  • Multi-file code understanding and implementation planning.
  • Higher reliability on patch generation, bug fixes, and refactors.
  • Better adherence to coding constraints (tests, style, architecture notes).
  • Strong performance in tool-augmented development loops.
  • Useful behavior for review tasks: summarizing diffs, identifying risks, and proposing safer alternatives.

It is particularly effective when your system can provide high-quality context (tests, existing modules, and clear acceptance criteria).

Technical Details

OpenAI model documentation lists GPT-5-Codex with:

  • 400K context window.
  • 128K max output tokens.
  • Text and image input support with text output.
  • Full compatibility with tool use and structured orchestration in the modern API stack.

The model was launched as part of OpenAI’s Codex-focused engineering push and is intended for assistant-like and autonomous coding workflows. Teams usually get the best results when this model is paired with repository retrieval, test execution tools, and strict acceptance checks instead of unconstrained free-form prompting.

Pricing & Access

Public API pricing (per 1M tokens):

  • Input: $1.25
  • Output: $10.00

Access:

  • OpenAI API endpoints supporting GPT-5 family models.
  • Codex-centered workflows in OpenAI tooling and partner integrations.

For cost control, teams usually combine retrieval filters, shorter output constraints, and targeted model routing (for example, run easy lint tasks on cheaper models and reserve Codex for hard edits).

Best Use Cases

Choose GPT-5-Codex for:

  • Multi-step coding agents with test-run and fix loops.
  • PR review copilots that need nuanced technical feedback.
  • Migration and refactor projects across large repositories.
  • Code synthesis from specs where deterministic output format matters.

It is less ideal for pure conversational tasks where coding specialization does not add value. For mixed workloads, a two-tier route is common: GPT-5 for general assistant turns and GPT-5-Codex only when code-change depth or review precision is required.

Comparisons

  • GPT-5 (OpenAI): GPT-5 is broader and often enough for mixed workloads; GPT-5-Codex usually wins on deep software-engineering tasks.
  • Claude Opus 4.6 (Anthropic): Both are top-tier coding choices; Codex often has stronger OpenAI-native toolchain ergonomics.
  • Gemini 3 Pro Preview (Google): Gemini is highly capable in multimodal and long-context workflows; GPT-5-Codex is often preferred for patch reliability in code-first pipelines.