TLDR: 1M context, soon to be 2M 2.5 series are all thinking models 2.5-Pro is the one released, exceptional performance across the board except factQA (beaten by GPT4.5) all results are @pass=1, no voting etc. to artificially boost scores possibly was…