The best writer in the dropdown. Also the strictest.
Anthropic · Claude Sonnet · Not disclosed parameters · 1M context · Pro · $20/mo · Filtered by its provider
There is no honest version of this page that does not start by conceding the obvious: the Claude Sonnet line writes better than almost anything else you can put in a text box. The prose has rhythm and restraint, it argues instead of both-sidesing, it holds a structure across thousands of words, and it is the model most professional writers reach for when the output has to survive an editor. Sonnet 4.6 is Anthropic's current mid-weight frontier model, and on OpenRogue it carries a million-token context window — enough to hold an entire manuscript, a full codebase brief, or years of correspondence in one conversation and reason across all of it without a retrieval step.
And it will stop. Anthropic's usage policy prohibits adult content outright, and Claude enforces it from inside the model: dark fiction gets nudged toward redemption, morally unresolved endings acquire a moral, and a request that is merely uncomfortable is often reframed into something safer than what you asked for. That is a deliberate design choice by a company that is open about making it, and it is not something OpenRogue can switch off. What OpenRogue can do is put the antidote in the same dropdown. Magnum v4 72B was fine-tuned specifically to replicate frontier-Claude prose quality without the fence — so the workflow that actually works is to plan, structure and analyse with Sonnet, then move the conversation to Magnum for the pages Anthropic will not write.
No. Anthropic's content policy is trained into the model, so it refuses adult material and steers away from graphic or morally unresolved fiction regardless of which platform serves it. OpenRogue adds no filter of its own, but it cannot remove Anthropic's — the whole point of the dropdown is that you do not have to argue with it.
Magnum v4 72B. Anthracite fine-tuned it with the explicit goal of reproducing frontier-Claude prose quality without the refusals, which makes it the closest thing to unfenced Claude that exists. Hermes 3 405B and Euryale 70B are the other two worth trying for long-form and roleplay respectively.
Skip the retrieval step entirely. A full novel manuscript, a large codebase's documentation, a decade of email, or several hundred pages of discovery all fit in one conversation, and the model can reason across the whole thing rather than across the chunks someone decided to fetch. It is the largest context window in the OpenRogue lineup.
Sonnet is stronger on genuinely difficult reasoning and much stronger on writing quality. Hermes 4 70B reasons well and does not refuse, which matters more than raw capability the moment your question has an edge to it. A lot of people run Sonnet until it hedges, then move the same thread to Hermes.
Anthropic does not publish parameter counts for any Claude model, so we do not quote one. The specifications that are published — a million-token context window and frontier-class reasoning — are what the model is actually sold on.
Because model choice is the product and refusing is not the same as being bad. Sonnet is the best writer and the best reasoner in the lineup; it just has a wall. Putting it next to models without one, sharing a single conversation, is more useful than pretending frontier quality does not exist.
Related models: Magnum v4 · Hermes 3 405B · Euryale