Dev AIOriginally published on Zhihu Pin, 2026-01-14
For the first time, I subscribed to an Ultra/Max tier AI service. Claude Max 5x, paid via the Apple Store, at 125 USD per month.
Previously, I had been using Codex / Claude Code / Cursor together. However, I found that my main driver was still Claude Code. Codex produces good results, but the wait is too long. Cursor gets routed once you burn through the 20 USD quota, and I feel it is better suited for when I am fine-tuning a project. Claude Code with Opus 4.5 combines both quality and speed for a very smooth experience; the only downside is that the Pro tier usage limit is too low.
After waiting out the 5-hour limit twice, I decided to subscribe to Max for a while to see if my efficiency can scale further as my token usage scales. (This was a view I formed a while back, during a period when I was heavily using AI for self-reflection, project management, and fitness/diet/investment management.) Plus, it coincided with the release of Claude Cowork, so I get to try that out as well.
Current AI Coding Toolchain
Comparison of Primary Tools
Comparison of Primary Tools| Tool | Role |
|---|---|
| Opus 4.5 | Primary driver for AI Coding |
| Codex 5.2 xhigh | Complex problems |
| Gemini 3.0 pro | Web dev / Aesthetic & styling design |
| Gemini 3 flash | Quick queries (the experience has been really good lately, very fast) |
| Sonnet 4.5 | Occasional use |
Last month's ccusage: $82.80, which went into building two small projects. This month, I'm aiming to build an open-source project on context engineering, a few small projects, and a blog post about Agent Skills.
Issues with Gemini 3 pro
Issues with Gemini 3 proThere is really something off with Gemini 3 pro. It might be due to a sliding context window, which makes its performance in certain scenarios even worse than 2.5 pro. It is the strongest at one-shot tasks, but the further the conversation goes, the more likely it is to fall apart completely. If only my company would let me use Opus 4.5 TAT