# 腾讯混元发布Hy3量化版本，Emad Mostaque称赞1-bit性能

- 来源：Emad (@EMostaque)
- 发布时间：2026-07-15 02:42
- AIHOT 分数：47
- AIHOT 链接：https://aihot.virxact.com/items/cmrl20is8006fbicqssynlu10
- 原文链接：https://x.com/EMostaque/status/2077101354254811542

## AI 摘要

腾讯混元发布旗舰级295B参数模型Hy3的1-bit与4-bit量化版本，可在单GPU上运行。Emad Mostaque引用该消息并指出，Hy3在1-bit量化下SWE-bench Verified得分75.4%、SWE-bench Pro得分53.9%，性能惊人。模型仅88GB，可在Macbook Max上运行。他预期基于NVP4训练的模型性能下降更少，并建议未来可全面采用二进制推理。用户可通过llama.cpp运行Hy3并启用MTP。

## 正文

75.4% SWE Bench Verified / 53.9% SWE Bench Pro on 1 bit quantisation is 🤪

This is in line with my expectations &amp； you can expect even lower drop off with NVP4 base trained models - why not run everything binary？

88 Gb so works on a Macbook Max

### 引用推文

> Tencent Hy：We've just released the 1-bit & 4-bit version of Hy3, a flagship-scale 295B model that can be served on a single GPU. 👌 Run Hy3 with llama.cpp, enable MTP, and...