LatitudeGames/Equinox-31B

Equinox-31B: Balanced Creative Storytelling

Equinox-31B, developed by LatitudeGames, is a 31 billion parameter model built upon Gemma 4 31B Instruct. Its unique strength lies in its balanced fine-tuning approach, blending two distinct creative directions to offer a versatile narrative experience.

Key Capabilities & Training

• Dual Narrative Focus: Equinox was SFT-trained using a dataset that combines "Wayfarer 2" data (dark, consequence-driven text adventures) and "Hearthfire" data (longform, slice-of-life, character-depth storytelling). This allows it to navigate both perilous scenarios and nuanced emotional interactions.
• Second-Person Present Tense: The primary training data emphasizes second-person present tense, making it ideal for interactive fiction and immersive role-playing experiences.
• Gemma 4 Foundation: Leveraging the Gemma 4 31B Instruct base, Equinox benefits from a robust language understanding and generation capability.

Use Cases & Limitations

• Versatile Story Generation: Ideal for developers seeking a model that can generate narratives balancing intense action with detailed character development and atmospheric scenes.
• Interactive Fiction: Particularly well-suited for text-based adventure games and role-playing scenarios where a balanced tone is desired.
• Not for Reasoning: It's important to note that no reasoning datasets were included during training, so it is not optimized for complex logical tasks.
• Specialized Alternatives: For maximally grim or purely atmospheric slow-burn narratives, LatitudeGames' other specialized models like Harbinger or Hearthfire might be more appropriate.