There are so many new models coming out, it's easy to forget about the old standbys. Deepseek/deepseek-v3.2 is still one of the most knowledgeable open source LLMs, and it's killer feature is the price.
On Openrouter is costs $0.26/M input tokens $0.38/M output tokens. That's an absolute steal of a deal for what you can do with it. It's a lot slower than some of the Flash/Lite models for agentic work, but for writing code, doing research, and answering questions, it's hard to beat for the money.