Magnum v4 72B is a 72.7 billion parameter language model built on top of the Qwen2.5-72B-Instruct base, designed specifically to emulate the nuanced prose quality associated with Claude the listed price Sonnet and Opus. The model is fully fine-tuned rather than adapted with adapters, and its training was carried out on 8x MI300X GPUs using curated conversation datasets such as anthracite-org/c2 logs 32k llama3 qwen2 v1.2, anthracite-org/kalo-opus-instruct-22k-no-refusal, and anthracite-org/nopm claude writing fixed, all prepared in the ChatML format that the model continues to expect at inference time. Because it inherits an instruct-tuned lineage, it is well suited to multi-turn chat, structured instruction following, and long-context work, with documentation pointing to an extended context length of roughly 131,072 tokens for handling lengthy documents and sustained interactions.

In practical terms, Magnum v4 72B is aimed at users who want rich, human-like writing from an open-weights model: creative storytelling, role-play, dialogue-heavy applications, and other tasks where stylistic quality matters more than narrow factual lookup. The combination of a large parameter count, a Claude-inspired fine-tuning objective, and ChatML prompting gives it a distinctive voice while still allowing system prompts and developer instructions to steer behavior. It integrates through OpenAI-compatible APIs, making it straightforward to drop into existing chat pipelines, and its open-weight release makes it attractive for self-hosting teams that need a writing-focused model without depending on a closed provider.