Composer 2.5 regression + premium models burning credits for nothing.
Composer 2.5 has become noticeably worse. The quality regression is frustrating, especially for Android Studio and PHP development.
The bigger issue is that premium models consume your usage like crazy without actually solving anything. Gemini, even the latest models, has been garbage in my experience. Grok 4.6/4.7 are equally disappointing for my workflow.
A single task using Gemini 3.8 Flash consumed 70% of my $60 plan without fixing the issue. The only models that have consistently delivered for me are ChatGPT 5.6 Sol and Claude.
We need better model quality, more consistent results, and usage limits that make sense. Paying $60/month only to burn through the allowance on failed attempts defeats the whole purpose of using AI for development.