Abliteration AI, a startup led by Devon Thomas, launched an OpenAI- and Anthropic-compatible hosted API around August 29-31, 2026. Its “abliterated-model-large-v2” is based on GLM-5.3, a 753-billion-parameter mixture-of-experts model whose weights were modified to reduce refusals for offensive tasks. Customers cannot access the modified weights and must use the hosted service, which features a one-million-token context window and is priced at $3 per million input tokens and $5 per million output tokens. The company reports self-assessed results of 84.5% on CyberGym, 41.8% on Terminal-Bench 4.0 and 105 tasks completed within two hours on ExploitGym. The service enforces a zero-data-retention policy while retaining safeguards against child sexual exploitation and self-harm, and other providers have already begun reselling similar GLM-5.3 variants.
Source: Read the original article
