The storytelling classic that started a lot of people's habit.
Gryphe · Llama 2 13B (MythoLogic-L2 × Huginn merge) · 13B parameters · 4K context · Free · Censorship: none
MythoMax is the model that taught a generation of people what an unfiltered AI actually feels like. Gryphe built it by merging MythoLogic-L2 with Huginn at the tensor level rather than fine-tuning from scratch, and the result had a quality nobody quite expected from 13B: warmth. It wrote characters that felt like they wanted something, it took a premise and ran with it instead of asking permission, and it did all of that on hardware a hobbyist could actually afford. For a long stretch it was the default answer to 'which model should I start with' on every roleplay forum, every self-hosting subreddit, every list of models worth quantizing. If you have ever used an AI character site that felt suspiciously good for its size, there is a decent chance MythoMax was underneath it.
It is an old model now and we are not going to pretend otherwise. The 4K context window is the smallest in the OpenRogue lineup by a wide margin — roughly a long scene before it starts forgetting the beginning — and its world knowledge stopped updating in the Llama 2 era. What survived the years is character: it still writes with a generosity that a lot of newer, smarter models have had trained out of them. That is why it sits on the free tier here. Start a scene with MythoMax to feel the difference between a filtered assistant and a model that simply plays along, then move up to Rocinante or Euryale when the story outgrows four thousand tokens.
For short scenes, character work, and quick fiction — yes, and the reason is style rather than intelligence. It writes with a warmth that a lot of newer models have had polished away, and at 13B it answers almost instantly. For anything that needs long memory or current world knowledge, pick a 70B from the lineup instead.
Yes. It predates the era when open models shipped with heavy refusal training, and it was merged from components chosen for creative willingness. It follows a scene where you take it, including adult and dark themes, without inserting a lecture. OpenRogue adds no filter layer on top of it.
MythoMax is built on Llama 2, whose native context is 4,096 tokens — that was the ceiling for the whole generation. Practically it means roughly one long scene or a few thousand words of history before the earliest turns start dropping out. Euryale 70B and Hermes 3 70B both give you 131K if memory is what you need.
Rocinante 12B is the modern equivalent: newer base, eight times the context, sharper plotting. MythoMax has the older, softer voice that a lot of people still prefer for intimate two-character scenes. Both are cheap to try, and OpenRogue lets you switch between them mid-conversation without losing the thread.
It is a merge, not a fine-tune. Gryphe combined MythoLogic-L2 and Huginn using a layered tensor blend, taking the reasoning behaviour from one parent and the prose behaviour from the other. That technique is routine now; MythoMax is one of the models that proved it worked.
Yes — it sits on OpenRogue's free tier, so you can start a scene without a card. The free tier has a daily message cap; any paid plan lifts it and unlocks the larger uncensored models alongside it.