Input COBOL files
AI Models & Assistants / Knowledge library
GPT-5.3-Codex
The most capable agentic coding model to date.
Why learn GPT-5.3-Codex?
Offers the documented OpenAI model capabilities through hosted.
The problem it solves
Provides model inference for supported text / image inputs.
Where it came from
GPT-5.3-Codex: The most capable agentic coding model to date. Published aliases: gpt-5.3-codex. Input: text, image. Output: text. Checked 2026-10-13. Model capabilities are distinct from tools available in a particular agent.
Build your understanding
Work through these exercises in order. Try the task before opening a reference.
- Explain what GPT-5.3-Codex does.
- Identify a use case for GPT-5.3-Codex.
- Describe a limitation of GPT-5.3-Codex.
- Name a compatible tool for GPT-5.3-Codex.
- Compare GPT-5.3-Codex against an alternative.
Documentation and sources
- Primary source / model card (developers.openai.com)
Catalog research notes
- See the AI Workbench for source dates, pricing evidence and configuration details.
Connect the ideas
Built on
Real World Example
Automated Legacy Financial Refactoring
Global investment banks maintaining monolithic COBOL-based ledger systems.
The business problem
The high cost and extreme risk associated with manual refactoring of decades-old, undocumented financial logic during cloud migration projects.
How this topic solves it
Utilizing GPT-5.3-Codex via the Responses API to perform agentic code analysis and transformation. By leveraging the model's high reasoning effort settings and hosted_shell tool, the agent can iteratively analyze legacy source files, generate unit tests, and propose refactored modules in modern languages while maintaining functional parity.
- Ingest legacy source code and documentation into the model context using the 400,000 token window.
- Configure the agent with high reasoning effort to map complex business logic dependencies.
- Execute iterative refactoring cycles using the hosted_shell tool to validate compilation and logic integrity.
- Generate structured outputs for the refactored code blocks to ensure integration with modern CI/CD pipelines.
The flow from legacy code ingestion to validated modern output.
Agentic reasoning engine
Validation environment
Refactored output
- Legacy Source → Ingests context → GPT-5.3-Codex
- GPT-5.3-Codex → Executes tests → Hosted Shell
- Hosted Shell → Returns results → GPT-5.3-Codex
- GPT-5.3-Codex → Outputs code → Modern Module
Expected business outcome
A significantly accelerated migration timeline with improved code maintainability and reduced regression risks compared to manual conversion.
Limitations and checks
The model may occasionally hallucinate edge-case business rules in highly obfuscated code; verify by checking the output against the original logic using the model's function_calling capability to cross-reference against existing test suites.