the grok-3 benchmark is pretty useful in comparing base models, so I added GPT-4.5
https://x.com/bycloudai/status/1895233722309599364?s=46