Omartificial-Intelligence-Space/ara-jev-Faisal

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 28, 2026License:cc-by-nc-4.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Faisal is a 4.5 billion parameter Arabic-first decision model developed by Omartificial-Intelligence-Space, built on Qwen3.5-4B. It specializes in classification and decision-making tasks by taking a context, a question, and 2-24 options, then selecting one with calibrated probabilities. This model excels at applying policies written in plain Arabic, understanding customer intent across dialects, and routing tickets, making it ideal for automated decision systems in Arabic-speaking contexts.

Loading preview...

Faisal: An Arabic-First Decision Model

Faisal is a 4.5 billion parameter model developed by Omartificial-Intelligence-Space, specifically designed for Arabic-language decision-making and classification tasks. Built upon Qwen/Qwen3.5-4B, it functions as an "arbiter" that evaluates a given context and question against 2 to 24 provided options, returning a single choice with calibrated probabilities for each option. This model is optimized for speed and accuracy in Arabic environments.

Key Capabilities

  • Arabic Policy Enforcement: Achieves 100% accuracy on held-out prose policies and 99.3% on policies from unseen domains, effectively applying rules written in plain Arabic.
  • Contextual Understanding: Can process customer intent, even with dialectal variations (Gulf, Egyptian, Levantine), and handles code-switching.
  • Robustness: Treats input text as data, ignoring embedded instructions or prompt injection attempts within the context.
  • Calibrated Confidence: Provides probability scores for each option, enabling reliable auto-acceptance (98.1% accuracy for p > 0.99) and human review splits.
  • Flexible Options: Accepts 2-24 options at request time, allowing for dynamic classification without retraining.
  • Efficient Decision-Making: Utilizes a single forward pass to score valid option letters, always returning exactly one choice without free-form generation.

Good for

  • Automated Customer Support: Classifying customer intent, routing tickets, and analyzing sentiment in Arabic.
  • Policy Adherence: Automatically applying written policies to request data or other structured/unstructured inputs.
  • Content Moderation: Categorizing news, identifying offensive language, or performing other classification tasks in Arabic.
  • Any Arabic-first decision task where a clear choice needs to be made from a set of options with high confidence.

Finetunes
1 models