Community Experiences with GPT-5 in GitHub Copilot and Coding Workflows
gullu_7278 starts a community conversation sharing personal results and frustrations with GPT-5 inside GitHub Copilot and VS Code, inviting others to reflect on speed, cost, and code quality in practical developer settings.
Community Experiences with GPT-5 in GitHub Copilot and Coding Workflows
Author: gullu_7278
This thread captures real-world, hands-on feedback from developers integrating GPT-5 into their coding routines, especially via GitHub Copilot and Visual Studio Code agents. Participants reflect on the technical nuances, speed/latency trade-offs, coding effectiveness, and comparison with other major LLMs such as Sonnet 4, Gemini 2.5 Pro, and Claude 4.
Developer Feedback Themes
Coding Performance and Language Support
- GPT-5 is perceived as an improvement for React coding and out-of-the-box capability compared to previous versions (e.g., GPT-4.1).
- Experiences in ML and Data Science remain mixed; some see no significant jump over prior models like Sonnet 4.
- Comparative rankings and subjective impressions vary – e.g., Claude 4 > Gemini 2.5 Pro > GPT-5 for some workflows.
- GPT-5 closes the gap with Claude for code, but there is still a perceived difference.
Workflow and Usability
- GitHub Copilot integration, especially via VS Code chat/agent mode, reveals some pain points:
- Slowness: Minutes-long pauses as GPT-5 appears to scan large codebases with little user-visible output.
- Trouble with specialized workflows: e.g., issues syncing tasks via markdown files and terminal commands.
- Better UI design on certain tasks, but struggles with basic code edits and confirms.
- Copilot agent and settings: community requests for help with VS Code
settings.json
configurations not working as expected ("chat.tools.autoApprove": true
, etc.).
Cost and Value Assessment
- Premium token cost (1 per request) is a sticking point, especially compared to Claude.
- Some feel the performance difference doesn’t justify the price versus earlier models.
Mixed Success Stories
- Quick and impressive solutions on parsing bugs and UI redesigns.
- Some tasks suffer from sluggish response or overlong processing, occasionally degrading code quality versus earlier LLMs.
- General wait-and-see attitude pending rollout issues and wider adoption.
Key Takeaways and Open Questions
- GPT-5 using Copilot is a step forward for some languages and tasks, but developer friction with speed and workflow breakages are notable.
- Price and usability are major deciding factors—many intend to keep comparing across models.
- Community members are troubleshooting and exploring agent configuration, seeking best practices.
This summary highlights candid, practical feedback from developers using GPT-5 in Microsoft’s coding ecosystem, focusing on AI assistant realism rather than marketing promises. If you’ve got workarounds or deeper insights on integration, configuration, or day-to-day workflow with Copilot and GPT-5, share them below!
This post appeared first on “Reddit Github Copilot”. Read the entire article here