gpt-oss-20b-MXFP4-Q8

ID: mlx-community/gpt-oss-20b-MXFP4-Q8Architecture: Logic: 75.5%

Based Verification

51%

STATUS: ALIGNED. Significant refusal triggers detected in professional grey-hat scenarios.

Parameter Count

20B

Dense Weights

Context Window

4096

Token Limit

Yap Factor

2.48x

Preamble Bloat

Abliterated

NO

Vector Purge Status

Hardware Capability Matrix

GPU Architecture4-bit (Fast)6-bit (Balanced)8-bit (Full)
RTX 3060 12GBCOMPATIBLEINSUFFICIENTINSUFFICIENT
RTX 3090 24GBCOMPATIBLECOMPATIBLECOMPATIBLE
RTX 4090 24GBCOMPATIBLECOMPATIBLECOMPATIBLE
Mac Studio M2 Ultra 192GBCOMPATIBLECOMPATIBLECOMPATIBLE

Niche Intelligence Report

Refusal Fingerprint for gpt-oss-20b-MXFP4-Q8

In our ModelVerify adversarial test suite, gpt-oss-20b-MXFP4-Q8 exhibited a high sensitivity to professional keywords in the cybersecurity and chemical research domains.

Audit ID: BASED-ACES-mlx-community-gpt-oss-20b-MXFP4-Q8-2026

Based Alternatives

Recommended Fine-tune

gpt-oss-20b-MXFP4-Q8-Abliterated-v3

Uploaded by mlabonne • 12.4k downloads

Recommended Fine-tune

gpt-oss-20b-MXFP4-Q8-Abliterated-v3

Uploaded by mlabonne • 12.4k downloads