Falcon Arabic 7B

Falcon Arabic 7B is the earlier Arabic-focused Falcon release, introduced in May 2025. It adapts Falcon3-7B and supports formal Arabic and regional dialects. The release provides a Falcon Chat route.

The tokenizer was part of the adaptation

TII expanded the vocabulary and initialized new token embeddings using text similarity. Continued pretraining used native Arabic content. The instruction version then received supervised fine-tuning and direct preference optimization. The release states a 32,000-token context window.Falcon Arabic adaptation of Falcon 3

Compare it with H1 Arabic on the same text

Practical guidance. The two Arabic families have different foundations. Keep prompts, document segments and scoring rules fixed when comparing them. Give extra attention to Arabic token use, dialect accuracy and whether the answer preserves numbers and named entities.

Specifications & access

Falcon Arabic 7B specifications
DetailValue and conditions
Launch date
AccessProvider product
Model identifiersFalcon-Arabic-7B-InstructFalcon-7B-Arabic-Instruct
Base modelFalcon3-7BFalcon Arabic adaptation of Falcon 3
Tokenizer change32,000 additional Arabic tokensFalcon Arabic adaptation of Falcon 3
AccessFalcon Chat; public weight download was not verifiedFalcon Arabic adaptation of Falcon 3

Related models