Stable Beluga

Stable Beluga is a powerful Large Language Model (LLM) renowned for its exceptional reasoning abilities across a wide range of benchmarks. This model is adept at handling diverse applications, from copywriting and answering general or scientific questions to generating creative ideas for events such as parties. With its advanced capabilities, Stable Beluga offers users an invaluable resource for enhancing their creative and analytical tasks. Whether you’re crafting compelling content, seeking detailed explanations for complex queries, or brainstorming unique concepts for social gatherings, Stable Beluga provides the insights and creativity needed to excel in a variety of scenarios.

Download Stable Beluga

Download Stable Beluga
You can download Stable Beluga for free and with just one click below. Start using one of the top artificial intelligence Language models today.
Download Stable Beluga

What is Stable Beluga?

Stable Beluga comprises two distinct models, Stable Beluga 1 and Stable Beluga 2, which are built on the foundations of the LLaMA models. These LLMs have been fine-tuned to enhance their reasoning capabilities and to perform a wide range of tasks with high accuracy.

Stable Beluga Features

Feature Stable Beluga 1 Stable Beluga 2
Model Foundation LLaMA 65B LLaMA 2 70B
Fine-Tuning Technique SFT (Alpaca Format) SFT (Alpaca Format)
Data Source Variation High-Quality Diverse Datasets High-Quality Diverse Datasets

Stable Beluga Performance

Performance Metric Stable Beluga 2 Stable Beluga 1
Leaderboard Ranking 1st Place (Open LLM Leaderboard) 4th Place (Open LLM Leaderboard)
Benchmark Tools EleutherAI’s lm-eval-harness, AGIEval EleutherAI’s lm-eval-harness, AGIEval

Stable Beluga Capabilities

Specialized Domain Knowledge: They excel in fields requiring high expertise, such as legal and mathematical contexts.
Linguistic Nuances: The models’ ability to understand and manipulate language is finely tuned, allowing them to handle complex queries and generate nuanced responses.
Innovative Applications: With their open-access nature and robust performance, the models are ideal for developing new AI applications across educational, scientific, and creative fields.
Stable Beluga Technical Details
  • Foundation Models: Stable Beluga 1 is based on LLaMA 65B, and Stable Beluga 2 on LLaMA 2 70B.
  • Fine-Tuning Technique: Both models utilize Supervised Fine-Tune (SFT) in the standard Alpaca format, employing a synthetically-generated dataset for enhanced training efficacy.
  • Data Source Variation: Unlike other models, Stable Beluga’s training data comes from high-quality, diverse datasets, ensuring a broad understanding of various subjects and scenarios.
Stable Beluga Performance
  • Leaderboard Rankings: As of the latest evaluations, Stable Beluga 2 ranks first on the Open LLM Leaderboard, with Stable Beluga 1 not far behind at fourth.
  • Benchmarking Tools: The models’ capabilities were rigorously tested using tools like EleutherAI’s lm-eval-harness and AGIEval, confirming their top-tier performance.



Stable Beluga 1 and Stable Beluga 2 represent a leap forward in the development of LLMs, combining high-level technical capabilities with ethical and accessible research practices. Their introduction not only enhances the AI research landscape but also paves the way for innovative applications that can benefit a wide range of industries. With their release, Stability AI continues to demonstrate its commitment to advancing AI technology while fostering an open and collaborative research environment.