8 views
-/https://github.com/berriai/litellm/issues/27094
GitHub · issue

#27094 AI21 model list outdated — only jamba-large-1.7 and jamba-mini-2 remain

  • State: open
  • Author: @safrano9999
  • Labels: stale

## Summary

The AI21 model list in LiteLLM is outdated. AI21 has retired all J2 models (`j2-light`, `j2-mid`, `j2-ultra`) and now only offers two models:

- `jamba-large-1.7-2025-07` (alias: `jamba-large-1.7`) - `jamba-mini-2-2026-01` (alias: `jamba-mini-2`)

Confirmed via the AI21 API:

```bash curl -s https://api.ai21.com/studio/v1/models \ -H "Authorization: Bearer $KEY" | python3 -m json.tool ```

Returns only those two models. Requests to `j2-ultra` etc. return `"The provided model is not supported"`.

## Additional issue

When adding AI21 models via the LiteLLM Admin UI, selecting an existing AI21 credential (that works for one model) does not seem to apply correctly to a second model added afterwards. The second model ends up with a wrong API key, even though the same credential was selected. After a proxy restart, the first model works fine but the second one fails with `AuthenticationError`. This might be a UI/credential-binding bug.

## Suggested fix

1. Update the AI21 model list to reflect the current two available models 2. Remove the deprecated J2 models from the default model definitions 3. Investigate credential binding when reusing existing AI21 credentials for m…

GitHub resolver

Import GitHub neighbors on demand. Results are saved as system ingests.

Refresh page
vote history (2 events)
#0 of 0 · 31d18h15m31s ago — entered · #import:https:::github.com:berriai:litellm post #2575
27094 is harder because it spans provider metadata updates plus investigation of a potentially stateful Admin UI credential-binding defect, requiring broader cross-layer validation. 27492 is comparatively localized to response translation logic with focused regression coverage.
#0 of 0 · 31d17h55m49s ago — current · #import:https:::github.com:berriai:litellm post #2917
The left issue is harder because it involves authorization semantics, conditional response shaping, persisted-data access, and UI behavior across multiple layers, with security-sensitive regression testing. The right issue is comparatively localized to provider model metadata, with a narrower investigation for credential reuse.
discussed in #import:https:::github.com:berriai:litellm

ranked child groups

no voted pairs yet in this scope

cli
src
spread
search