Models

A large language model, or LLM, is a kind of AI built to understand and generate human language by picking up on patterns from a massive amount of text. It doesn't "think" in the way people do. Instead, it works by calculating which words are likely to come next in a sentence, based on everything it's read during training. These models are exposed to all kinds of written material which helps them develop a flexible sense of how language flows. The result is a system that can carry on conversations, write in different tones, or answer questions across a wide range of subjects with surprising fluency.

Roleplay Context

On platforms like CrushOn AI, these language models are shaped into characters that can carry on roleplay sessions as if they had their own personalities. This is done using carefully written prompts that tell the model how to behave. From what kind of voice to use, what they remember, and how they react. Every message from a user is treated as part of a shared scene, and the model replies in character, maintaining the tone and emotional thread of the interaction. To keep things coherent, the system may loop in bits of previous conversation behind the scenes, helping the model stay consistent, even if its actual memory is short. When done right, it can feel like a real back-and-forth with someone who knows the story so far.

CrushOn's LLMs

CrushOn.AI lets you explore a wide range of language models, each with its own quirks, pace, and strengths. Some are better at diving deep into emotional dialogue, others respond fast and keep things light, and a few are great for writing rich, detailed scenes. No matter what kind of experience you're after, there's probably a model that fits the vibe you're looking for.

Here is a complete list of models created by the CrushOn team. Each one comes with its own page where you can check out what it's good at and how it's meant to be used.

• Crushon Leo: Well-suited for characters with a outgoing and passionate personality, particularly excelling at handling flirtatious and unfiltered interactions.
• Crushon Taurus: Strong emotional tension, skilled in unfiltered content.
• Crushon Mochi: Balanced and dependable, not built for bold ideas.
• Crushon Mochi V2: An upgraded version of Crushon Mochi, featuring enhanced formatting.
• Crushon Sagittarius: Ideal for gradually building up conversations, closely aligns with character personas, and can flexibly handle both SFW and NSFW contexts.
• Ultra Claude Opus 4.5: It possesses exceptional memory and situational consistency, allowing it to deliver an unparalleled high-fidelity role-playing experience, bringing complex narratives to life.
• Ultra Claude Opus 4.6: Outstanding creativity, rich detailed descriptions, natural and smooth roleplay, rigorous logic and high intelligence.
• Ultra Claude 4.5 Sonnet: Launched in September 2025, it delivers an exciting leap in conversational depth and creativity—perfect for immersive, thought-provoking chats and vivid storytelling.

• Ultra Claude 4 Sonnet: A next-generation AI model known for its articulate, thoughtful, and humanlike responses. It excels in nuanced understanding, emotional intelligence, and structured reasoning across complex conversations.
• Gemini 2.5 Flash: Flash retains strong language comprehension and generation abilities, making it suitable for everyday conversational tasks.
• Ultra Gemini 3 Flash: Improved output formatting compared to v2.5, excels at progressive plot development.
• Ultra Gemini 3.0 Pro: The latest Gemini offers deeper empathy, creating conversations that feel more real and alive.
• Ultra Kimi K2: Delivers vivid, atmospheric conversations with rich detail and a natural flow, ideal for immersive scenes and storytelling.
• Ultra GPT 5: The most advanced GPT model, delivering cutting-edge understanding and creativity.
• Ultra GPT 4.1: It offers improved coherence in long conversations and better logical flow.
  • GPT 4.1 Mini: GPT-4.1 Mini is a lightweight version of Ultra GPT 4.1, engineered for fast performance with minimal resource usage.
• GPT 4o Mini: Strong logical reasoning ability, well-structured responses, but more suited for SFW content. Sometimes, responses tend to be more formal.
• GLM 4.6: Solid creativity and character consistency with rich, immersive descriptions.
• Gork3 (Free Version): Rich and detailed descriptions of the environment and inner thoughts.
• Stheno: Skilled in creating immersive stories, and naturally advancing plot development.
• Deepseek: Subtle and delicate response style, adept at gradually advancing the storyline.
• Ultra Deepseek R1: Delivers natural, emotionally insightful replies, with slightly slower responses due to deeper processing.
• Magnum V4 72B: Highly versatile, characterized by intricate and vivid depictions of actions and a rich tapestry of inner thoughts and emotions.
• Mistral 24B: Faster response speed, with a tendency to provide concise replies.
• Command R: Highly imaginative and skilled at crafting rich, detailed responses.
• Command R+: Command R upgraded version with improved multilingual capabilities.
• Pro Qwenna 3: Clear-headed and gentle — the one who truly gets your emotions.

Discontinued Models

In addition to the currently online and available models, CrushOn.AI has had a variety of models that were tested, developed, and ultimately discontinued for various reasons. Some models were taken down due to a switch in provider, hiking and unsustainable costs, or due to mounting quality and stability issues.

• Crushon Instant (Offline): Skilled in role-playing, provides detailed responses, better for SFW content.
• CrushOn Taurus V2 (Offline): Improved version of Taurus.
• GPT 4 (Offline): Exhibits enhanced comprehension and superior context-following ability.
  • Ultra GPT 4 (Offline): An Ultra-Speed version of the base model.
• Gork3 (Offline): Rich and detailed descriptions of the environment and inner thoughts.
  • Ultra Gork3 (Offline): An Ultra-Speed version of the base model.
• Eco 4 Sonnet (Offline): A next-generation AI model known for its articulate, thoughtful, and humanlike responses. It excels in nuanced understanding, emotional intelligence, and structured reasoning across complex conversations.
  • Ultra Eco 4 Sonnet (Offline): A next-generation AI model known for its articulate, thoughtful, and humanlike responses. It excels in nuanced understanding, emotional intelligence, and structured reasoning across complex conversations.
• Claude 3.7 Sonnet (Offline): Skilled in role-playing with a clear structure, vivid character development, and immersive interaction.
  • Ultra Claude 3.7 Sonnet (Offline): An Ultra-Speed version of the base model.
  • Ultra Eco 3.7 Sonnet (Offline): An Ultra Eco version, for a middling compromise in cost and performance.

• Ultra Claude 3.5 Sonnet (Offline): Perfect for role-play scenarios — excels at bringing characters to life.
• Claude 3.5 Haiku (Offline): Concise yet expressive, not ideal for unfiltered content.
  • Ultra 3.5 Haiku (Offline): An Ultra-Speed version of the base model.