February 26, 2026
It seems that whenever Anthropic, Google or OpenAI1 release a new model then everyone immediately makes it their new default. I think for most use cases the previous generation of models continue to perform adequately and that models from outside the major three providers are overlooked entirely.
I’ve paid for an LLM for a grand total of one time2. This isn’t to boast, it’s that I’ve never felt that it was necessary. The vast majority of my own usage of LLMs has been on the web with ChatGPT or Claude directly, thus I’ve practically never been on the cutting edge of the latest model release anyway, and I don’t think I’ve suffered at all because of it. The difference in capability for everyday tasks is rarely meaningful.
The ever so popular OpenCode project3 always has models available that are completely free to use and that offer a chance to stay in the loop. They’ve been the ones to drive my own exploration of models from other providers, like GLM-4.7, MiniMax M2.5, and Grok Code Fast 1. What initially started as a desire to use a genuinely good agent harness4 now means that I’m more often than not using models that are outmoded, but are still more capable than people give them credit for. As an N of 1 example: I was able to migrate both my Cometary and Wishlist Lite tools to newer major versions of Charm libraries with just a single prompt5.
A caveat though: free models, like OpenCode’s Big Pickle, can be slow to use and can involve noticeable queue times for requests, which is definitely in stark contrast to the big three, but I would still implore people to give other models a shot6.
To me the list of major providers is Anthropic, Google, and OpenAI. I’m leaving out xAI because as of late their models don’t seem to get anywhere near the level of attraction as the big three. ↩
And this was somewhat indirectly through Lovable to get a side project moving far quicker than I could without it. ↩
At least going by its GitHub star history. ↩
I would not be at all surprised if they get acquired as they seem to be in a league of their own. They offer both a TUI and a more traditional GUI application. In my opinion both are fantastic to use. ↩
Even managed to add a new feature to Cometary that I’ve been thinking about for the longest time. ↩
I’m still using other code search tools, like Sourcegraph, to learn how other projects are doing things on top of only relying on LLMs. ↩