I bought their $200 a year plan, their Pro plan. Anthropic's Claude has come a long way. 3.5 was super good, but Gemini 2.5 Pro March 2025 dwarfed that model. Unmistakably, Gemini was the best hands on, but their regression with recent Gemini models caused me to re-evaluate Claude. Claude's up to 4.5 well, 4.6 Opus (Maybe I'll try it) and Sonnet is really really good.
I've blogged before about my frustration with Gemini on how it just jumps to coding while in discussion, it's terrible and very bad regression. If Google climbs back to Gemini 2.5 Pro March 2025 tier, then I'll pay for that.
OpenAI is iterating too, but they are expensive relatively. I don't think GPT one shots things enough to justify token costs. OpenAI and Gemini are focused on multi-modal models. Claude could do multi modal, but nah screw that, they are focused on being the best LLM with the best software engineering assist tools. I think that right there is the selling Claude to Software Engineers.
I'll be making use of Claude like I did with Gemini 2.5 Pro March 2025 and seeing how I can leverage it to do hard work and offload easier work to local LLMs. Suffice to say, on the local LLM front, NVIDIA's Nemotron 3 is really good. GLM 4.7 Flash and Open AI's open models are good too, but i like NVIDIA's more.
Claude is just better at following directions and discussion, this is the most important characteristic I desire from an LLM, that's why I choose it. It doesn't force solutions down my throat, I can discuss with it and then we can move on to specifications and then planning and execution. That's my development flow, that's what I did with Gemini 2.5 Pro March 2025 and I'm surprised to say Claude, after about a year of Gemini, is where its at.