This is a quick rundown of the AI tools I'm currently using day-to-day.
Work
Claude Code Team Max 5x
This is my main tool for coding and automation. There's been a lot of talk about people switching to Codex because tokens burn out too fast or performance has dropped — but running Opus 4.7 with 1M context at medium effort all day for work, I've barely hit the 5-hour weekly limit, so I'll probably stick with this for a while.
I've occasionally used Claude Cowork for research, or kicked off parallel research sessions inside Claude Code. The limit did kick in a few times — Deep Research seems to chew through tokens at a noticeably higher rate.
For plugins, I'm mainly using Oh My Claude Code (OMC). There are strong competitors out there like Superpowers and others, but OMC hasn't given me any real reason to switch yet.
That said, I think Claude will eventually absorb the key features of these plugins natively, so I'm starting to question whether it's worth hopping between them. Probably better to stick with what I have and spend that energy actually building things — refining my own skills and harness.
GitHub Copilot Business
This was already in place before Claude Code came along. At this point it's basically sitting around as a backup — I hook it up through Opencode whenever Claude Code hits its limit. Starting in June, Microsoft is switching from a per-call limit to a token-consumption model, which honestly makes it feel even less useful as a fallback.
On the subject of Opencode — it's an open-source AI coding agent with 168k stars on GitHub and over 900 contributors. It's a seriously impressive project. Sometimes I actually prefer its agentic flow over Claude Code. I've also installed Oh-my-opencode, a plugin built by a young Korean developer, and it works well.
If I could connect Anthropic's models to Opencode under a subscription plan, I might honestly be using Opencode instead of Claude Code. The thing is, Anthropic has blocked plan-based usage of their models in Opencode — which is a competing product — while quietly porting ideas from Opencode and Oh-my-opencode into Claude Code. So really, I don't have much choice but to stay on Claude Code.
That said, if the big three — Anthropic, OpenAI, and Google — keep pushing prices up, while Chinese models continue to undercut them aggressively, connecting Kimi or GLM to Opencode becomes a pretty viable alternative. Running a local LLM is another option on the table, but until models in the 16–24GB range can deliver genuinely practical performance, it's not really production-ready for day-to-day work.
Personal
Claude Code Pro
Using the same Claude stack as at work makes sense for continuity, and since Sonnet 4.6 I haven't had many complaints. Also using OMC here. The one real pain point is the 5-hour limit — it's pretty easy to hit on personal projects, so I've been half-considering moving to Codex as my primary personal tool. Recent user reports suggest that 1) Codex performance is on par with Claude — some even say it handles complex logic problems better — and 2) you get significantly more tokens on the same $20 plan.
GitHub Copilot Pro
Whenever Claude Code hits its limit, I fire up Opencode and connect Copilot to keep going. It's also useful for getting exposure to different models. But with the upcoming billing changes, it's starting to lose its appeal — time to think about alternatives.
Z.ai GLM Coding Lite (monthly plan)
This is the GLM5 model I have connected as the primary in OpenClaw. Honestly, it's smarter than I expected — in the AI tools world, it's basically the BYD of the lineup: punching well above its weight for the price. Whether I end up using this as my GitHub Copilot replacement or eventually look at another Chinese LLM, I'll need a bit more time to decide. Back on GLM 4.7 the gap compared to Claude Sonnet was too wide to be practically useful, but since GLM 5.0 the performance has jumped noticeably — throw a reasonably complex task at it and it handles it well enough.
Wrapping Up
I'm still actively using all of the tools mentioned above. Claude Code remains the primary workhorse, but especially for personal use, cost is a real factor — so GitHub Copilot and GLM are filling in as supporting players.
I'd love to simplify everything and just go with Claude Max, but two things are holding me back: 1) the cost is a real stretch, and 2) it can't be used through OpenClaw or other automation tools. So for now, I'll keep rolling with this main AI + backup AI setup.