# Hugging Face — Quantization-Aware Healing: a compressed, 4-bit model that outperforms its full-precision original

- Company: Hugging Face (huggingface.co)
- Announced: 2026-08-25T11:39:24+00:00
- Category: research-paper
- Subject: Platform
- Models affected: GPT-OSS 120B
- Source: https://huggingface.co/blog/MultiverseComputingCAI/quantization-aware-healing
- Record: https://forck.live/items/4396-quantization-aware-healing-a-compressed-4-bit-model-that-outperforms-its-full

Introduces Quantization-Aware Healing (QAH), a method that distills from the original pre-compression model to recover capabilities lost during structural compression and quantization. Applied to a GPT-OSS 120B model compressed to 60B and quantized to MXFP4, it outperforms its own full-precision version on 7 of 9 benchmarks.

## Evidence

Verbatim from https://huggingface.co/blog/MultiverseComputingCAI/quantization-aware-healing:

> applied to a GPT-OSS 120B model compressed to 60B parameters and quantized to MXFP4, it produces a model that beats its own full-precision (bfloat16) version on 7 of 9 benchmarks.

---

Record: https://forck.live/items/4396-quantization-aware-healing-a-compressed-4-bit-model-that-outperforms-its-full
Catalogue: https://forck.live/llms.txt
Feed: https://forck.live/feed.md