Hello Everyone, I have been using both models in the Augment for the past few days, and here are my thoughts on it. Maybe it will help you too in some way.
Sonnet 4: More faster, more precise, better in UI, follows the guidelines and works superfine. (That is why it was the only model available in Augment.)
GPT 5: Better in terms of larger codebases, more good for complex backend tasks, heavy on tool calling, follow prompt instructions better.
It is good that now we have 2 options of top-of-the-line models in augment, it will be better for us only as it will lower the load on the API of particular models, and now we also have a model where we can fallback in case one model is unable to complete the task.
I have tried and tested GPT-5 on many AI IDEs and tools, but in my testing, it works the best with Augment Code only. Maybe due to the context engine of it.
Although there are still some issues in the GPT-5, I hope the OPENAI will make it more stable with time, and also the Augment Code will optimise itself based on the new model also as they did for Sonnet 4.
Thanks for that feedback. That is interesting. Does anyone else here have experience with this and think the same? Or, do you have another opinion on it?
Usually the guidelines for gpt5 only works for the Chat.. Doing so in agent only bypasses it then continues on a long loop of context engine reading and multiple reading of other codes in codebase.
One update, sir. The GPT 5 tends to now follow user guidelines at all. They follow the prompt fully (which sonnet) lag, but they skip the user guidelines completely. Please take a look to see if we can improve it.
I have to agree. I just spent 2 hours trying to complete simple tasks that would have been done by Claude in 30 minutes! it is so disappointing.
the UI outputs are pedestrian, it is not following my requests, and it takes forever to come back with an answer.
Gpt5 isn’t working for me. My workflow with large changes doesn’t work with gpt5. I have to look after the messages as I keep getting questions from gpt5. I don’t like it.
Since I can’t edit my original post, I’d like to add a few updates.
I spent several hours with GPT-5, and the experience was disappointing. Besides its slow response times, the results were consistently broken:
• Uploading a CSV file to Neon failed completely.
• The model didn’t recognize the connection string in my .env file.
• It couldn’t identify basic CLI tools or the MCP server.
I also asked it to create a new page in my app that connects to a Neon table, pulls the headers, and matches the look and functionality of the existing pages. Even after a lengthy back-and-forth, it still couldn’t follow my instructions.
What I found most frustrating was the lack of meaningful feedback or personality—it was almost impossible to understand its thought process.
I switched to Sonnet and, within 30–60 minutes, built two additional pages and added several features.
Overall, GPT-5 was a disappointment. I’d rather spend my money on Opus 4.1.
Are there any plans or an ETA for this implementation?