OpenAI Codex was a dedicated API for AI code generation based on GPT-3 and fine-tuned on public GitHub code. OpenAI shut it down on March 23, 2023, because newer models like GPT-3.5-Turbo and later successors outperformed it on coding tasks.
What It Was
Codex launched in August 2021 as a free beta API. It powered GitHub Copilot and supported code completion, generation from natural language, code explanation, unit test creation, and cross-language translation across languages like Python, JavaScript, and SQL.
The API offered models such as code-davinci-002 for high capability and code-cushman-001 for speed. Pricing started at $0.10 per 1,000 tokens upon general availability. GitHub Copilot handled costs via subscription for its users.
What Happened
OpenAI announced the Codex API deprecation on March 23, 2023, with models code-davinci-002, code-davinci-001, code-cushman-002, and code-cushman-001 going offline that day. Developers were directed to GPT-3.5-Turbo or GPT-4o as replacements, with no migration period.
GitHub Copilot transitioned to newer models around that time. A later codex-mini-latest model was deprecated on February 12, 2026, but the core Codex API ended in 2023.
Why It Died
Codex succeeded in powering tools like Copilot but became redundant when general-purpose models like GPT-3.5-Turbo and GPT-4 surpassed its coding performance. OpenAI prioritized investment in newer foundation models over maintaining a specialized API.
The sudden shutdown without warning disrupted production systems dependent on Codex models.
Current Alternatives
Current replacements for Codex API users include:
- GitHub Copilot, IDE-integrated coding assistant using latest OpenAI models
- Cursor, AI-powered code editor with frontier model support
- Aider, Open-source CLI tool for AI-assisted coding
- Claude Code, Anthropic’s coding interface via API and apps
- Codeium, Free AI coding autocomplete for multiple IDEs
Lessons
Specialized AI models like Codex often get replaced by advancing general-purpose foundation models, as seen with GPT-3.5 and beyond. Companies prioritize scalable, multi-task models over niche APIs.
Sudden deprecations without transition periods risk breaking user workflows; phased sunsets with clear timelines build trust.
Rapid AI progress shortens product lifecycles, requiring developers to build model-agnostic systems for resilience.