A lot changed in the last 18 months. Claude Opus 4.7. GPT-5.5. Llama 4 Maverick eating the open-weight market. Google I/O going all-in on agents. And cost collapsing 20-50x since GPT-4 launched.

    I wrote up a survey of where things stand — frontier models, open-source contenders, and the trends that actually matter. No predictions. Verified against official sources.

    https://proddraft.com/blog/state-of-llms-may-2026

    Next up: a real-world comparison — running Llama 4 locally vs. calling an API model for a production workload.