Qwen3.8-27B scored 29/30 on AIME 2026 with FP8 + xhigh reasoning — BF16 vs FP8 results
英文标题：Qwen3.8-27B scored 29/30 on AIME 2026 with FP8 + xhigh reasoning — BF16 vs FP8 results
I benchmarked Qwen3.8-27B on MathArena/aime_2026 dataset, comparing BF16 and FP8 weights at medium and xhigh reasoning effort. Interesting findings are: quantized FP8 xhigh is better than BF 16 medium equally good as 16 BF xhigh with better speed. On problem 7, both BF16 xhigh and quantized FP8 xhi
打开原文相关 2 · 机会 0
中文
I benchmarked Qwen3.8-27B on MathArena/aime_2026 dataset, comparing BF16 and FP8 weights at medium and xhigh reasoning effort. Interesting findings are: quantized FP8 xhigh is better than BF 16 medium equally good as 16 BF xhigh with better speed. On problem 7, both BF16 xhigh and quantized FP8 xhi
中文全文翻译正在补齐，可先查看下方英文原文。
English / 原文
I benchmarked Qwen3.8-27B on MathArena/aime_2026 dataset, comparing BF16 and FP8 weights at medium and xhigh reasoning effort. Interesting findings are: quantized FP8 xhigh is better than BF 16 medium equally good as 16 BF xhigh with better speed. On problem 7, both BF16 xhigh and quantized FP8 xhi