Yeah, at this point with these factors:
- Disappointing Grok 4.7 release - both in model quality/efficiency and with integration bugs
- Impressive open-weight releases of GLM-5.3 Flash, Deepseek V4.1 Flash, and MiMo V2.6-Pro - and the lack of Cursor integration for open-weight models at this cost-to-intelligence ratio since Deepseek V4 Flash 0731
- Impressive OpenAI and Anthropic releases for GPT-6 Sol/Luna and Claude Opus 5.5
- Upcoming OpenAI departure in November, and already current restriction of GPT-6 models
I’ve practically made up my mind that if we don’t get a sequel Composer model that can compete on value proposition by November, I’m 100% leaving to explore Codex, Claude Code, and other possible alternatives. The 1st-party usage limits in Cursor are incredibly generous, but considering I’m only allowed to spend it on Grok and Composer 2.5, that’s not exactly a great deal.
This sucks because I love Cursor’s harness and workflow, and I’ve had worse experiences where that’s specifically concerned on other platforms. But if it means getting more for my money, I’m willing to deal with that learning curve tradeoff.