More useful than many benchmarks: GPT 5.6 Sol just improved tensor parallel prefill numbers preserving numerical correctness in DwarfStar after many rounds of Fable optimizations. Does not mean it is better, bit it is a strong signal IMHO.

More useful than many benchmarks: GPT 5.6 Sol just improved tensor parallel prefill numbers preserving numerical correctness in DwarfStar after many rounds of Fable optimizations. Does not mean it is better, bit it is a strong signal IMHO.