Qwen 3.6 Plus is the new price/performance leader of all frontier offerings. Its knowledge rivals that of the newest GPT and Claude models, but it's priced at $0.325/M input tokens $1.95/M output tokens.
From my previous post on Rebolforum, some of the other recent best choices:
Google/gemini-3.1-flash-lite-preview is one of the best price/performance models currently available. It's not just smart, but also very knowledgeable - that's something most of the inexpensive fast models are missing. Most models in the 'flash' category, and other smaller models like Minimax, can perform tasks intelligently, but they don't have a ton of world knowledge or obscure info built into their parameters. This one is very fast, very cheap for its capability ($0.25/M input tokens, $1.50/M output tokens), and has a huge amount of built in knowledge. I think this will likely replace the current models I'm using in agentic systems.
Minimax is also exciting right now for coding (I mentioned that their hosted chat system is the only one besides Chat-GPT which I can use natively to perform my preferred software development workflow with zip files (this requires absolutely no local agent setup, which is huge for me)).
I'm also waiting eagerly to see if MiMo-V2-Pro gets open sourced.
Kimi 2.5 is still the best huge open source model, but GLM and Deepseek are very capable, and I can't wait to see Deepseek 4...