Jais 3 375B-A23B
Jais 3 375B-A23B is the flagship sparse release. The developer lists 375B stored parameters and 23B active per token. The advertised context window is 512K, and the model card links an FP8 version.
What the release includes
The published weights use Apache 2.0. The card targets Arabic and English assistant and agent tasks. It does not establish equivalent performance in other languages.Jais 3 375B-A23B official model card
Evaluate the complete serving system
Practical guidance. This is a capacity-planning decision as well as a model choice. Compare task quality, weight placement, prompt processing and concurrent throughput. The active parameter count cannot by itself predict the number of accelerators required.Jais 3 375B-A23B official model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | Jais-3-375B-A23B |
| Architecture | Mixture of experts; 375B total, 23B activeJais 3 375B-A23B official model card |
| Published context | 512K; developer notationJais 3 375B-A23B official model card |
| Weight license | Apache 2.0Jais 3 375B-A23B official model card |
Related models
- Jais 3 32B
Inception42 (G42) · Jais 3
Largest dense Jais 3 model, with 32 billion parameters and a stated 512K context window.
Launch date not confirmedDownloadable - Jais 3 MoVA 36B-A4B
Inception42 (G42) · Jais 3
Arabic-and-English mixture-of-experts model combining 36 billion stored parameters with Mixture-of-Values attention.
Launch date not confirmedDownloadable