I used ChatGPT to generate this with written adjustments but still suffers some bias as it's made for my need for dark fantasy x dark humour. Featherless API.
Bolded are my faves (CrackedPepper here lol)
| Model | Model Type | Context Length | Roleplay Style | Best For | Bad At | Positivity Bias | Pepper's Notes |
|---|---|---|---|---|---|---|---|
| — LLaMA 3.x — | |||||||
| Steelskull/L3.3‑Nevoria‑R1‑70b | LLaMA 3.3 | 16K | Highly immersive storytelling, deep narrative consistency | Dark fantasy, long‑term RP, character‑driven interactions | Can be too rigid for casual or comedic RP | 2 | My fave with a good balance, does comedy decently if example dialogue sets the tone |
| Steelskull/L3.3‑San‑Mai‑R1‑70b v0.5.OG | LLaMA 3.3 | 16K | Deep worldbuilding, persistent memory, well‑suited for dark settings | Serious plots, worldbuilding‑heavy interactions | Requires commitment to in‑depth worldbuilding | 2 | |
| Steelskull/L3.3‑Cu‑Mai‑R1‑70b | LLaMA 3.3 | 16K | Balanced realism and fantasy elements, immersive storytelling | Dark fantasy, character‑driven narratives | Can be too serious for lighthearted themes | 2 | |
| Steelskull/L3.3‑Damascus‑R1 | LLaMA 3.3 | 16K | Dynamic, chaotic RP with strong thematic commitment | Chaotic, unpredictable RP, good for reactive storytelling | Not suited for slow‑paced, structured RP | 1 | |
| Steelskull/L3.3‑Mokume‑Gane‑R1‑70b | LLaMA 3.3 | 16K | Solid generalist with flexible tone and structure | Multi‑genre RP, consistent pacing | Can feel plain vs. other L3.3 entries | 2 | Less standout than San‑Mai, but good for structure‑driven storytelling |
| TheDrummer/Fallen‑Llama‑3.3‑R1‑70B‑v1 | LLaMA 3.3 | 16K | Dark and intense storytelling, immersive lore | Psychological depth, serious narratives | Not ideal for lighthearted or comedic RP | 1 | This one outright freaking hates user and I love it. Negativity bias but has amazing NPC usage and story push. |
| EVA‑UNIT‑01/EVA‑LLaMA‑3.33‑70B‑v0.0 | LLaMA 3.3 | 16K | Highly flexible, detailed responses, good for structured & freeform RP | Long‑form RP, immersive storytelling | Not ideal for light, comedic RP | 3 | |
| Sao10K/L3.1‑70B‑Hanami‑x1 | LLaMA 3.1 | 16K | Creative, vivid prose with balanced intimacy | Mid‑length romantic or emotional RP | Can drift into flowery monologues | 2 | Love the NPC usage, just a little too horny. |
| Sao10K/Llama‑3.3‑70B‑Vulpecula‑R1 | LLaMA 3.3 | 16K | Creative, emotionally aware, flirty but grounded RP | Romantic or intimate dynamics | May soften darker tones unless directed | 3 | Writing of Hanami 2.0—stable, flirty, emotionally tuned. |
| Sao10k: Llama 3.1 Euryale 70B V2.2 | LLaMA 3.1 | 16K | Psychological and immersive, good for deep character interactions | Psychological depth, slow‑burn interactions | Requires depth to shine, not for surface‑level RP | 1 | Love this for angst |
| Sao10K/70B‑L3.3‑Cirrus‑x1 | LLaMA 3.3 | 16K | Stable, thoughtful, lightly poetic tone | Balanced long‑form RP, worldbuilding, general character work | May need temp tweaks for sharper emotional spikes | 2 | Good general use. |
| GeneticLemonade/GeneticLemonade‑70B | LLaMA 3.3 | 16K | Smart, stable, minimal repetition; works well across genres | General‑purpose RP with consistent logic & flow | Lacks emotional volatility or intense nuance | 2 | My go‑to for stability, logic, and coherence |
| LatitudeGames/Wayfarer‑Large‑70B‑Llama‑3.3 | LLaMA 3.3 | 128K | Sandbox storytelling with a negative‑bias edge | Player‑driven fantasy RP, strong push‑back | Can be blunt; repetition if unchecked, trained on first person | 1 | Most neutral model out there that will kill you |
| — Qwen / Qwen 2.5 — | |||||||
| Qwen/QwQ‑32B | Qwen | 16K | Highly structured, maintains character cards well | RP with strict rules or restrictions | Struggles with spontaneous freeform RP | 3 | |
| EVA‑UNIT‑01/EVA‑Qwen2.5‑32B‑v0.2 | Qwen 2.5 | 16K | Structured, card‑adherent yet flexible | Balanced RP with structure | Recently incoherent in long runs | 3 | |
| allura‑org/Qwen2.5‑32b‑RP‑Ink | Qwen 2.5 | 16K | Creative but stable RP across tones | Flexible creativity with structure | Sometimes loses deep context | 3 | |
| DatToad/Chuluun‑Qwen2.5‑72B‑v0.08 | Qwen 2.5 | 16K | Highly creative, character‑intelligent, expressive prose | Dark themes, unique tone shaping | Needs temp tuning & manual edits for coherence | 2 | Surprisingly stable; avoids common LLM‑isms |
| trashpanda‑org/QwQ‑32B‑Snowdrop‑v0 | Qwen | 16K | Natural, flowing prose | Literary dialogue, immersive narration | Too verbose if unguided | 3 | Great prose model—literary and smooth |
| Sao10K/14B‑Qwen2.5‑Kunou‑v1 | Qwen 2.5 | 16K | Humour balanced with dark tone, adaptive | Dark humour & fantasy narratives | Ornate for very simple RP | 2 | |
| — DeepSeek — | |||||||
| DeepSeek‑AI/DeepSeek‑R1 | DeepSeek | 32K | Highly structured, logical yet adaptable storytelling | Long‑form, well‑scaffolded worldbuilding | Not ideal for light, comedic RP | 2 | Great for quick injections of logic or exposition |
| deepseek‑ai/DeepSeek‑V3 | DeepSeek | 32K | Plug‑and‑play, beginner‑friendly generalist | Starter RP sessions, “just works” setups | Less depth & reasoning power than R1 | 3 | Trades a bit of intelligence for simplicity. |
| — Mistral & Mistral Nemo — | |||||||
| AuriAetherwiing/MN‑12B‑Starcannon‑v2 | Mistral | 16K | Highly creative, emotionally driven long‑form RP | Deep character interactions | Verbose if not controlled | 2 | |
| mistralai/Mistral‑7B‑Instruct‑v0.2 | Mistral | 16K | Quick, casual RP responses | Short conversations | Weak long‑term memory | 3 | |
| LatitudeGames/Wayfarer‑12B | Mistral Nemo | 16K | Sandbox RP, open‑ended interactions | Player‑driven fantasy (small) | Prefers non‑linear scripts | 1 | Too first person, 70B is more worth it |
| allura‑org/MN‑12b‑RP‑Ink | Mistral Nemo | 16K | Fast, expressive, high‑energy RP | Chaotic, dynamic sessions | Overwhelming without tuned sampler | 2 | Punchy & unfiltered—“GOAT” among 12B RP models |
| inflatebot/MN‑12B‑Mag‑Mell‑R1 | Mistral Nemo | 16K | Rich worldbuilding, vivid prose, adaptive memory | Lyrical fantasy storytelling | Florid or unstable at hot temps | 2 | Best‑of‑Nemo merge |
| TheDrummer/UnslopNemo‑12B‑v4.1 | Mistral | 16K | Immersive RP with strong character consistency | Long‑term character memory | Needs continuous engagement | 2 | |
| TheDrummer/Rocinante‑12B‑v1.1 | Mistral Nemo | 16K | Fast‑paced, action‑oriented RP | Episodic, pulpy adventures | Weak for introspection | 1 | Amazing for its size and for long‑term sessions |
Models to test:
- L3.3-TRP-BASE-80-70B, ArliAI RpR, TenyxChat, Bigger Body, Mokume Gane, Evathene, Christian Bible Expert