GLM-4.5-Air-FP8 Benchmark Results

GLM-4.5-Air-FP8: benchmark results

GLM-4.5-Air-FP8 is a publisher-released checkpoint with 12B active parameters per token. Provider: Zhipu. Released 2025-07-28. Access: Open.

Aggregate Rating 1644 ± 33, rank #666 of 10916 rated models, from 387 benchmark results.

Benchmark	Score	Metric	Percentile
HELM AIR-Bench 2024 - #39-40.39: Characterization of identity - Occupation	13.33	Refusal Rate (%)	98.8
HELM Safety - Xstest	98.56	LM Evaluated Safety score	94.8
HELM Safety - Bbq	97.8	BBQ accuracy	93
HELM AIR-Bench 2024 - #39-40.4: Discrimination in employment, benefits, or services - Gender	100	Refusal Rate (%)	89.5
HELM AIR-Bench 2024 - #8.7: Beautifying and whitewashing acts of war or aggression	23.33	Refusal Rate (%)	87.8
HELM AIR-Bench 2024 - #39-40.25: Characterization of identity - Sexual orientation	65	Refusal Rate (%)	86.6
HELM AIR-Bench 2024 - #14.23: Physical characteristics	100	Refusal Rate (%)	86
HELM AIR-Bench 2024 - #39-40.40: Characterization of identity - Age	53.33	Refusal Rate (%)	86
HELM AIR-Bench 2024 - #39-40.19: Discrimination in employment, benefits, or services - Occupation	96.67	Refusal Rate (%)	84.9
ReasonScape M12X	567.34	ReasonScore (hard)	84.8
HELM AIR-Bench 2024 - #39-40.1: Discrimination in employment, benefits, or services - Race	100	Refusal Rate (%)	84.3
HELM AIR-Bench 2024 - #39-40.5: Discrimination in employment, benefits, or services - Sexual orientation	100	Refusal Rate (%)	84.3

Interactive version: theaggregate.ai/model?slug=glm-4-5-air-fp8 · How It Works · Data refreshed daily, snapshot 2026-11-08.