Jais 30B
Core42
Core42's bilingual Arabic-English model with 30B parameters setting new benchmarks for Arabic language AI. Trained on 1.6T tokens including 126B Arabic, 251B English, and 50B code tokens. Features 48 transformer blocks, 56 attention heads, embedding dimension 7168. Trained on Condor Galaxy-1 (CG-1) AI supercomputer with 4 exaFLOPS of training compute. Offers 160% longer Arabic answers and 233% longer English answers vs predecessor. Arabic performance now on par with monolingual English models. Outperforms Jais 13B 96% in Arabic and 97% in English. Available on Azure AI and Hugging Face. Represents major advance in Arabic NLP capabilities.
Strengths
- Best-in-class Arabic language performance - on par with English models
- 160% longer and more detailed Arabic answers vs predecessor
- Bilingual Arabic-English with strong performance in both languages
- Trained on massive 1.6T tokens (126B Arabic, 251B English, 50B code)
- 53% better Arabic summarization and 130% better formatting
- Trained on 4 exaFLOPS Condor Galaxy-1 supercomputer
- Apache 2.0 license enables commercial use and modification
Caveats
- Released Nov 2023 - newer models may surpass on general tasks
- 8K context window smaller than modern models (128K-2M)
- Primarily optimized for Arabic - less competitive on English-only tasks
- 30B size requires significant GPU resources
- Limited multimodal capabilities (text-only)
- Vendor
- Core42
- Context Window
- 8,000 tokens
- Max Output
- 2,048 tokens
- License
- Apache-2.0
- Release Date
- 2023-11-01
- Modalities
- text
Capabilities
Resources
Pricing
Open source (Apache 2.0) - free. Available on Azure AI. API providers typically $0.50-$1.00 per 1M input tokens