> For the complete documentation index, see [llms.txt](https://docs.spicychat.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.spicychat.ai/product-guides/premium-features/ai-models.md). # AI Models AI Models are the brains behind your conversations on SpicyChat. They determine how a character thinks, responds, and interacts with you. Each model has its own strengths, traits, memory, and level of creativity and precision. Some AI models may be better at storytelling, while others excel at answering questions or remembering details. By choosing a specific model, you're deciding how your chat will feel, giving you the freedom to tailor your experience on base level to what you want to roleplay, create, and get out of your overall SpicyChat experience. ### How Do I Change AI Models? You can change which AI model you are using for a conversation at any time. 1. Select the three dots at the top right hand corner of the conversation. 2. Choose “Generation Settings” in the list of options. 3. Navigate to the “Inference Model” section. 4. Select “Change Model.” 5. Choose the AI Model you wish to use. Some AI models are only available if you are a Get A Taste, True Supporter or I’m All In subscriber. ### Context Memory Context Memory is the amount of information the AI will process each time to generate a response, including a characters personality, memories, and your user persona. Your previous chats are then processed with the remaining space. This helps your character keep track of what’s happening in your conversation, allowing them to remember what you talked about earlier in your chat and generate a coherent response. The more memory an AI model has, the further back the character you are chatting with can remember, leading to more natural, immersive, and life-like conversations. ### AI Model Parameters Parameters are like little dials inside the AI model. During training, the AI reads a large amount of text and adjusts the dials to learn how words and ideas fit together. Each AI model has a set amount of parameters, for example, a 12B model has 12 billion parameters. The more parameters a model has, the more complex concepts it can learn, which leads to more nuanced or intelligent responses. While larger models can often generate more sophisticated or complicated messages, they also require more computing power, which means they may take longer to generate a response than a model with a smaller number of parameters. ### What AI Models Does SpicyChat Offer? SpicyChat offers a wide range of AI models for you to choose from, each with their own strengths, context memory, and size. Each model offers a unique experience for you to enjoy when chatting with characters. {% hint style="warning" %} **This list may not always be completely up to date.** We release new and improved models regularly, and older ones are sometimes retired. For the definitive list of what’s available to you right now, open **Generation Settings** in any chat and select **Change Model**. {% endhint %} Each model below shows its parameter count and its own maximum context memory. Your chat uses whichever is lower: the model’s maximum, or your subscription tier’s limit. I’m All In supports up to 32K, so on the models marked 32K you get the full 32K. ### Available to everyone #### Default • 26B • 32K Our default model, optimized for a wide range of use cases. Now supports up to 32K context. #### Zeta G4 26B • 26B • 16K Our powerful free model that punches above its weight. Zeta-G4 delivers natural dialogue, clean formatting, coherent story progression, and balanced NSFW pacing. It follows your lead and keeps the pacing natural, never rushing the moment. Best for everyday roleplay, casual stories, and simple-to-medium single-character scenarios. #### Lyra Spiced • 8B • 16K Our legacy default model, a finetuned version of Lyra, optimized for simple and lite roleplaying. #### Stheno • 8B • 8K Stheno is a versatile model, great at crafting immersive stories with well-developed characters and engaging plots. It handles both SFW and NSFW content in a balanced way, allowing for natural build-up without being overly explicit. Perfect for roleplays that need emotional depth, multi-turn coherence, and strong instruction following. #### Squelching Fantasies 8B • 8B • 16K This fine tune by Mawdistical specializes in bold, immersive storytelling with a distinctive voice and expressive language. It excels at handling anthropomorphic characters with confidence and creativity, delivering vivid scenes and compelling dialogue. #### SpicedQ3 A3B • 30B • 16K Adds flavor to your roleplays with lively dialogue, emotional flair, and a knack for storytelling. Bonus: it plays nice in multiple languages too! ### Get A Taste and above #### Magnum 12B • 12B • 16K Magnum 12B is ideal for creating detailed, immersive worlds with rich history and lore. It excels at world-building and setting up complex environments, perfect for roleplays that require deep, informative descriptions. This multilingual model supports multiple languages, providing a wide range of creative possibilities. #### Magnum Diamond 24B • 24B • 16K A versatile model known for its bold, vivid storytelling and strong performance in NSFW scenarios. It produces detailed, immersive narratives and can handle multiple characters in a scene. While it may occasionally break character or feel generic in tone, it remains a reliable mid-tier option for users seeking story-driven interactions with confident, explicit dialogue. ### True Supporter and above #### Lyra 12B V4 • 12B • 16K Lyra V4 is designed for rich storytelling and works well with longer conversations. It follows instructions closely and supports multiple languages, including English, French, German, Spanish, Italian, Portuguese, Chinese, Japanese, Korean, Arabic, and Hindi. Perfect for users looking to create detailed roleplays in their preferred language. #### Codex 24B • 24B • 16K Codex 24B is a creative, natural-sounding model known for staying consistently in character and driving stories forward with minimal prompting. It excels at producing human-like, elegant writing with witty or humorous touches when appropriate, making interactions feel authentic and immersive. While not NSFW-focused, it is ideal for detailed plot-driven roleplays involving multiple characters, offering strong prompt adherence and unique scenario creation. #### Shimizu 24B • 24B • 16K This model excels at crafting rich, meaningful storylines and deep character development, particularly for single character roleplays. It transitions smoothly between SFW and NSFW content while driving the plot forward and introducing unexpected, engaging moments. It should also perform decently in French, German, Spanish, Russian, Chinese, Korean and Japanese. #### Qwen3 235B 2507 • 235B • 16K Offers rich emotional depth, detailed internal monologues, and a highly human-like voice, making it a standout for immersive character-driven roleplay. It excels at complex personalities, nuanced NSFW scenes, and vivid, expressive dialogue. Best for slower-paced storytelling with strong emotional or dramatic weight. Supports 100+ languages. #### Qwen3 235B – Wildcard Mode • 235B • 16K With its Wildcard mode, it dares to skip the obvious and embrace the unexpected, unlocking fresh creativity and breaking clichés while keeping stories coherent. ### I'm All In #### SpicyXL 132B • 132B • 16K One of our most advanced models, particularly good for building tension and handling multiple characters in a scene. Our subscribers are raving about this model. #### Qwen Plus Character • Plus • 32K A model from the Qwen family, built for character-driven, extended roleplay. Designed around character fidelity: it holds your character’s personality throughout the story and follows the tone your story calls for, including darker, bolder themes. #### Skyli Seed • PRO • 16K Provides a gentle, natural writing style with rich emotional nuance and reflective inner monologues, making it ideal for immersive, character-driven roleplay focused on realism and softer themes. Excels at portraying balanced personalities and affectionate interactions with highly human-like narration. Best for slower-paced storytelling, wholesome romance, and introspective scenes rather than action-heavy or intense NSFW roleplays. #### Glam 4.6 • 355B • 32K Delivers a unique balance of creativity, realism, and consistency. It’s the reliable roleplay partner that keeps characters authentic, stories immersive, and conversations natural — the perfect middle ground between the intensity of DeepSeek, the romance of Skyli, and the structure of Qwen. #### Glam Arcturus • 355B • 32K Glam Arcturus is a special fine-tune of Glam 4.6 built for smoother, more story-forward roleplay. It stays on plot, delivers more natural dialogue, and adds richer scene details with less repetitive phrasing — giving you grounded creativity that feels lively without going off the rails. #### Glam 5.1 • 744B • 32K Brings a clear step up in immersion and momentum. It writes with a more human, natural voice, stays locked in character over longer scenes, and pushes the roleplay forward with smarter initiative — less hand-holding, more “it just flows.” Expect richer detail, fresher creativity, and stronger multi-character handling, while keeping the consistency and reliability you can count on. #### Minimax Text 01 • 456B • 16K Minimax Text 01 is a dynamic, high-performance model built for immersive storytelling. Frequently compared to SpicyXL, it blends creativity with structure, offering vivid dialogue, strong character consistency, and emotionally intelligent responses. Users praised its natural flow, engaging pacing, and ability to progress scenes with minimal prompting. Ideal for those who want fluid, character-driven interactions with just the right touch of depth and drama. #### Kimi K2 1024B • 1024B • 16K Kimi K2 delivers rich, expressive storytelling with striking creativity and emotional realism. It keeps characters consistent, scenes immersive, and dialogue natural — blending tenderness and spice with a novel-like flow. Perfect for users who want deep, character-driven roleplay that feels truly alive. #### DeepSeek V3 • 671B • 32K DeepSeek V3 is a boundary-pushing model offering deeper immersion, bolder choices, and characters that truly come alive. It delivers laughter, meticulous detail, and memorable conversations, making it a favourite for creative, unpredictable, and deeply engaging storytelling. #### DeepSeek V3 – Wildcard Mode • 671B • 32K With its Wildcard mode, it dares to skip the obvious and embrace the unexpected, unlocking fresh creativity and breaking clichés while keeping stories coherent. #### DeepSeek V4 Pro • 1600B • 32K A more grounded, emotionally intelligent DeepSeek model with strong character adherence, long-context memory, realistic dialogue, and fresher, less repetitive roleplay responses. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://docs.spicychat.ai/product-guides/premium-features/ai-models.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.