Jamba 1.5a
Jamba 1.5a is a documented research adaptation of Jamba Mini 1.5. AI21 and Enkrypt AI used additional preference training to adjust the model's behavior to a code of conduct.
Access not confirmed
What the extra training changed
The project used synthetic preference data, direct preference optimization and rejection sampling. Its goal was to change helpfulness, harmlessness and honesty while preserving the underlying model's capabilities.AI21 and Enkrypt AI's Jamba 1.5a research
A model adaptation rather than a separate filter
The white paper describes applying a DPO variant to the model weights with Enkrypt AI's synthetic alignment pipeline. The study explored whether an outside organization could adapt an already released model to its own conduct rules.Jamba 1.5a alignment white paper
Research evidence does not establish access
Practical guidance. The publication identifies a trained derivative, but it does not establish a current general-access endpoint or downloadable checkpoint. The underlying Mini 1.5 weights remain a separate release.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Access not confirmed |
| Launch date | |
| Model identifiers | Jamba-1.5a |
| Base model | Jamba Mini 1.5Jamba 1.5a alignment white paper |
| Collaborator | Enkrypt AIAI21 and Enkrypt AI's Jamba 1.5a research |
| Distribution | Research is documented; a public weight or API route is not confirmedAI21 and Enkrypt AI's Jamba 1.5a research |
Related models
- Jamba Tiny Dev
AI21 Labs · Jamba research
319-million-parameter trained development checkpoint for work on the Jamba architecture.
Launch date Research releaseDownloadable - Jamba Mini 1.5
AI21 Labs · Jamba
Original instruction-tuned Mini model with tool calls, JSON output and long text input.
Launch date Deprecated API; check access detailsDownloadable · API