FP8 quantization

Qwen 27B Ties Claude Opus on AIME 2026; Open-Source LLMs Trail GPT by 3 Points

Alibaba's Qwen 27B scored 96.7% on AIME 2026 (29/30), tying Claude Opus 4.6 but trailing GPT-5.6's 99.9% by 3 points. FP8 runs 2.7x faster.