OpenAI Unveils GPT-5-Codex in GitHub Copilot Public Preview: Next-Generation Agentic Coding Transforms Developer Productivity
Published: 24 September 2025 | AI Development, GitHub Copilot, OpenAI
OpenAI announced on 23 September 2025 that GPT-5-Codex, its next-generation coding model optimised for agentic development, is now rolling out in public preview for GitHub Copilot. Available to Copilot Pro, Pro+, Business, and Enterprise subscribers, this advanced model represents a significant leap in autonomous coding capabilities, enabling developers to tackle complex multi-step challenges through natural conversation.
GPT-5-Codex: Optimised for Agentic Coding Workflows
From Code Completion to Autonomous Development
GPT-5-Codex marks OpenAI's strategic shift from traditional code completion to fully autonomous software development. The model is specifically optimised for "agentic coding" – the ability to understand complex requirements, break them into sub-tasks, and execute comprehensive solutions independently.
Key Capabilities:
- Agentic Reasoning: Advanced multi-step problem decomposition and solution planning
- Contextual Understanding: Deep comprehension of codebases and architectural patterns
- Autonomous Execution: Independent implementation of complex features from high-level requirements
- Iterative Refinement: Continuous improvement through testing and validation cycles
- Production-Ready Output: Enterprise-quality code with proper error handling and testing
Model Picker Integration in VS Code
GitHub Copilot users can now select GPT-5-Codex directly from the model picker within Visual Studio Code. The model is available in three operational modes tailored to different development workflows:
Available Modes:
- Ask Mode: Interactive Q&A for technical guidance and code explanations
- Edit Mode: Direct code modifications based on natural language instructions
- Agent Mode: Fully autonomous feature implementation and problem-solving
This flexibility enables developers to choose the appropriate level of AI autonomy for each task, from quick questions to complete feature development.
Evolution from Original Codex to GPT-5-Codex
Historical Context: The Codex Journey
The original OpenAI Codex models, which powered early versions of GitHub Copilot, were standalone API products focused primarily on code completion and generation. These models have been deprecated as OpenAI transitioned to more advanced chat-based architectures.
Timeline of Evolution:
- 2021: Original Codex launch as standalone API
- 2024: Deprecation of Codex model family
- 2025: Introduction of new "Codex" agent powered by o3/o4-mini reasoning models
- September 2025: GPT-5-Codex public preview with agentic capabilities
What Makes GPT-5-Codex Different
GPT-5-Codex represents a fundamental architectural shift beyond simple code generation:
Technical Advancements:
- Reasoning Models: Built on OpenAI's latest o3/o4-mini reasoning architecture
- Agentic Workflows: Designed specifically for autonomous multi-step development
- Production Focus: Optimised for enterprise-quality code generation
- Multi-Model Integration: Part of GitHub Copilot's broader multi-model strategy
- Continuous Learning: Benefits from ongoing improvements to OpenAI's reasoning models
GitHub Copilot's Multi-Model Strategy
Provider Diversity Driving Innovation
GitHub's decision to support multiple AI providers reflects the rapid evolution of coding models. After opening support for multiple models in 2024, GitHub now ships new models "almost as fast as they drop" from providers including OpenAI, Anthropic, and Google.
Currently Supported Models:
- OpenAI: GPT-4o, GPT-5-Codex (new)
- Anthropic: Claude 3.5 Sonnet, Claude 3.5 Haiku
- Google: Gemini 2.0 Flash
- Provider-Specific: Various specialised models for different tasks
This multi-model approach enables GitHub Copilot to leverage the strengths of different AI architectures, providing developers with optimal tools for specific coding challenges.
Agent Mode: Autonomous Development at Scale
The introduction of "agent mode" on 6 February 2025 marked a significant milestone in GitHub Copilot's evolution. This mode enables truly autonomous operation, where Copilot independently implements features, creates tests, and handles edge cases without constant developer oversight.
Agent Mode Capabilities:
- Feature Implementation: Build complete features from high-level requirements
- Pull Request Reviews: Automatically review code by tagging the agent in PRs
- Bug Investigation: Autonomous debugging and fix implementation
- Test Generation: Comprehensive test suite creation
- Documentation: Automatic generation of technical documentation
Access and Availability
Subscription Requirements
GPT-5-Codex access is available exclusively to GitHub Copilot subscribers with specific plan tiers:
Eligible Plans:
- Copilot Pro: Individual developers
- Copilot Pro+: Enhanced individual access
- Copilot Business: Team and department licenses
- Copilot Enterprise: Organisation-wide deployment
Current Limitations
The initial public preview includes certain access restrictions:
Access Requirements:
- IDE: VS Code Insiders build required initially
- Rollout: Gradual availability to eligible subscribers
- Model Selection: Manual selection via model picker
- Preview Status: Features and performance subject to change
UI Refresh and Enhanced Features
Modernised Interface
Alongside GPT-5-Codex, GitHub Copilot received a comprehensive UI refresh focused on developer experience:
Interface Improvements:
- Modern Design: Cleaner, more intuitive interface
- Enhanced Commands: Expanded command palette with natural language processing
- Performance Optimisation: Faster response times and lower latency
- Bug Fixes: Resolution of long-standing user-reported issues
- Accessibility: Improved keyboard navigation and screen reader support
Code Review Integration
One of the most significant new features is seamless code review integration. Developers can now request code reviews simply by tagging the GitHub Copilot agent in pull requests.
Code Review Capabilities:
- Automated Analysis: Comprehensive code quality assessment
- Security Scanning: Identification of potential vulnerabilities
- Best Practices: Suggestions aligned with team coding standards
- Performance Review: Analysis of algorithmic efficiency
- Documentation Check: Verification of adequate code documentation
UK Developer Impact
Competitive Advantages for UK Teams
GPT-5-Codex availability through GitHub Copilot provides UK development teams with cutting-edge AI capabilities, enabling them to compete globally whilst maintaining high-quality standards.
Strategic Benefits:
- Productivity Multiplication: Significant acceleration of development workflows
- Skills Amplification: Junior developers leveraging senior-level AI guidance
- Quality Consistency: Automated adherence to coding best practices
- Time-to-Market: Faster delivery of new features and products
- Cost Efficiency: Reduced development costs without compromising quality
Financial Services and Healthcare Applications
UK sectors with stringent regulatory requirements are particularly positioned to benefit from GPT-5-Codex's production-ready code generation:
Sector-Specific Advantages:
- Financial Services: Automated compliance with FCA regulations
- Healthcare: GDPR and NHS Digital standards implementation
- Government: Accessibility and security requirements adherence
- Fintech: Rapid deployment of new financial products
- Legal Tech: Complex business logic implementation
The Competitive Landscape
AI Coding Market Dynamics
The GPT-5-Codex release intensifies competition in the AI coding assistance market, with major players including OpenAI, Anthropic (Claude Code), Google (Gemini), and Amazon (CodeWhisperer) all advancing autonomous coding capabilities.
Market Positioning:
- OpenAI/GitHub: Integration advantage through GitHub platform
- Anthropic: Focus on extended autonomous operation (30+ hours)
- Google: Emphasis on speed with Gemini 2.0 Flash
- Microsoft: Azure and VS Code ecosystem integration
This competitive environment drives rapid innovation, benefiting developers through continuously improving AI coding tools.
Getting Started with GPT-5-Codex
Implementation Steps
UK developers ready to explore GPT-5-Codex can begin immediately:
Quick Start Guide:
- Verify Subscription: Ensure eligible GitHub Copilot plan
- Install VS Code Insiders: Download latest Insiders build
- Enable GitHub Copilot: Activate extension if not already enabled
- Select GPT-5-Codex: Choose model from picker in ask, edit, or agent modes
- Start Coding: Begin with simple tasks and progressively increase complexity
Best Practices for Adoption
Maximise GPT-5-Codex value with structured adoption:
Recommended Approach:
- Start with Ask Mode: Build familiarity through Q&A interactions
- Progress to Edit Mode: Leverage for targeted code modifications
- Experiment with Agent Mode: Test autonomous capabilities on non-critical features
- Establish Review Processes: Maintain human oversight initially
- Document Patterns: Record effective prompting strategies
- Share Knowledge: Establish team best practices and learning resources
Conclusion: The Agentic Coding Revolution
OpenAI's launch of GPT-5-Codex in GitHub Copilot represents a significant milestone in AI-powered software development. The shift from code completion to fully autonomous agentic coding fundamentally transforms developer workflows, enabling small teams to accomplish what previously required large departments.
For UK developers, GPT-5-Codex availability through GitHub Copilot's familiar interface lowers the barrier to adopting cutting-edge AI coding capabilities. Whether building fintech applications, healthcare platforms, or government services, the combination of agentic reasoning and production-ready code generation enables UK teams to compete globally whilst maintaining high-quality standards.
The public preview period offers an opportunity to explore GPT-5-Codex capabilities, establish best practices, and prepare for broader organisational adoption. As agentic coding becomes standard practice, early adopters gain significant competitive advantages through accelerated development cycles and enhanced productivity.
Ready to Experience Agentic Coding?
GitHub Copilot Pro, Pro+, Business, and Enterprise subscribers can access GPT-5-Codex today through VS Code Insiders. Visit GitHub Copilot to get started.
