jan-hq/stealth-v1.2: A Privacy-Focused 7B LLM

jan-hq/stealth-v1.2 is a 7 billion parameter language model developed by Jan, designed for local, offline execution. With an 8192-token context length, this model prioritizes user privacy and data sovereignty, ensuring conversations remain confidential and on-device. It is compatible with Jan Desktop, an open-source, OpenAI-compatible local server, allowing developers to run the model on Mac, Windows, or Linux.

Key Capabilities & Performance

This model demonstrates solid performance across a range of benchmarks, as evaluated on the Open LLM Leaderboard:

• Average Score: 70.68
• AI2 Reasoning Challenge (25-Shot): 66.38
• HellaSwag (10-Shot): 86.14
• MMLU (5-Shot): 64.33
• TruthfulQA (0-shot): 54.23
• Winogrande (5-shot): 80.74
• GSM8k (5-shot): 72.25

These scores indicate its proficiency in reasoning, common sense, and language understanding tasks.

Ideal Use Cases

• Privacy-Sensitive Applications: Excellent for scenarios where data privacy and local processing are paramount, such as personal assistants or enterprise tools handling confidential information.
• Offline Environments: Suitable for use cases requiring AI capabilities without an internet connection.
• Developer Prototyping: Its OpenAI-compatible local server makes it easy for developers to integrate and test AI applications locally.