abliterated-model-large-v2 is positioned as the default large model for users who need the deepest reasoning the provider offers. It is derived from the open-weight GLM-5.3 base, then further abliterated and fine-tuned by Abliteration AI, which means the starting point is a publicly available model rather than a closed proprietary build, and the provider's own post-training is what shapes its behavior. The model is text-only, so it focuses purely on language-in, language-out tasks rather than image or video understanding, and is intended for harder reasoning and evaluation workloads where a long context and careful thinking matter more than multimodal breadth.
In practice, the model fits workloads that benefit from a very large working memory and adjustable depth of thought, because it operates in three reasoning effort modes (low, high, and max) and exposes a 1,000,000-token context window. Pricing is uniform across input and output at $5.00 per the cataloged API limit, with cached input reads billed at $0.50 per the cataloged API limit, which gives large-context applications a predictable cost profile as prefixes get reused. Compared with the smaller abliterated-model, which is multimodal and limited to a 256K context, this large variant trades multimodality for a much longer horizon and stronger reasoning depth, making it a better match for long-document analysis, multi-step agentic planning, and evaluation pipelines where the extra context budget and higher reasoning budgets are worth the premium over the general-purpose default.
