I’ve been continuing to use my Claude Max x20 plan. I haven’t hit any limits yet.
I’ve been quite impressed with Matt Pocock’s improve-codebase-architecture skill. I’ve used it on a couple of repositories where I’d erred on the side of vibing and the skill’s ability to clean things up and create a much more sensible codebase structure has been very helpful.
I’m intrigued by OpenAI’s symphony spec for agent orchestration. I’ve made a start at setting up a system along its lines for myself, but haven’t gotten it to where I could actually use it yet.
A note from a friend got me looking into MiniMax M2.7. Its quality-to-price ranking is quite impressive, nearly reaching Opus 4.7 performance at a much lower per-token API price. Though that comes at the cost of sending your code to a company in China.
According to the same website, publicly-available GPT and Gemini models are currently ahead of Claude in terms of quality. I don’t know what I’ll do with that information given at least so far Anthropic feels like a more trust-worthy company compared to Google or OpenAI. It’s kind of amazing how sketchy the frontier labs seem to have managed to become.