Abliterated GLM-5.3 Available in Commercial API

AbliterationAI has released abliterated-model-large-v2 — an “abliterated” version of GLM 5.3 with refusal directions removed from its weights, hosted in FP8, with a 1M-token context (text only) and a price of $5 per 1M tokens for both input and output.

Compiled by Sergey KostenchukPublished 2026-09-03Updated 2026-09-18

2026-09-03 Research OpenAI

🤖 Abliterated GLM-5.3 Available via API

AbliterationAI has released abliterated-model-large-v2 — a version of GLM 5.3 with refusal directions removed from its weights, responsible for refusals. The model is hosted in FP8, with a 1M-token context and a price of $5 per 1M tokens for both input and output.

🌍 The provider is turning uncensored access to a frontier model into a standard OpenAI-compatible endpoint with zero prompt retention, targeting penetration testing and red-teaming. The claimed 84.5% on CyberGym is a vendor self-measurement, and the company calls comparisons with competitors non-strict.

👤 Owners of an OpenAI-compatible client only need to change one line — model: abliterated-model-large-v2. The web sandbox offers a free trial credit, with plans ranging from $20 to $200 per month. Metrics should be verified on your own tasks before production use.

Source 1: https://abliteration.ai/blog/introducing-abliterated-model-large-v2 Source 2: https://abliteration.ai/pricing

Sources