Models Overview
Not Just Another Model Family
Jan Models aren’t general-purpose chatbots. Each model is trained for specific tasks that matter in daily work: search, analysis, creative writing, coding, research. They work together, each handling what it does best.
Current Training Status
| Model | Specialization | Size | Training Progress | Status |
|---|---|---|---|---|
| Jan-Search | Web search + synthesis | 7B | ████████░░ 82% | Testing phase |
| Jan-Write | Creative + technical writing | 13B | ████░░░░░░ 41% | Active training |
| Jan-Analyze | Data analysis + reasoning | 13B | ██░░░░░░░░ 23% | Dataset prep |
| Jan-Code | Code generation + debugging | 7B | ░░░░░░░░░░ 0% | Starting Jan 2025 |
Why Specialized Models?
One model can’t excel at everything. GPT-4o, o3, o4, or Claude 4 Sonnet writing poetry use the same weights to do math as well, which can be inefficient and expensive.
Our approach:
- Jan-Search knows how to query, crawl, and synthesize
- Jan-Write understands tone, structure, and creativity
- Jan-Analyze excels at reasoning and data interpretation
- Models work together through the Jan orchestration layer
Built for the Ecosystem
These aren’t standalone models. They’re designed to:
- Run efficiently on local hardware (quantized to 4-8GB)
- Work with Jan Tools (browser automation, file parsing, memory)
- Scale from laptop to server without code changes
- Share context and hand off tasks to each other
Help Us Improve
Models are only as good as their real-world performance. Test our models against your actual use cases and vote on what works.
We train on your feedback, not just benchmarks.