How to Explore the Falcon Models on falconllm.tii.ae

Updated

To explore the Falcon models, start at the model insight cards on the falconllm.tii.ae homepage and follow each card's "Learn More" link to the model's own page. The site is organized as a model catalog with announcement sections rather than a single download or pricing page, so the reliable path is: read the card summary, open the model page, then check the Latest Updates and Insights sections for newer releases. The page evidence does not show pricing, download mechanics, or login requirements, so treat those as things to confirm on each model page rather than assume.

What the site actually presents

The homepage introduces the Falcon LLM family from the Technology Innovation Institute and frames it around "Powering the Future of Responsible AI." It lists models as insight cards, each with a short description and a "Learn More" link. The visible cards cover:

Model What the card says it is for
Falcon Perception A multimodal model that lets systems "see, read, and understand images using natural language prompts," combining vision and language
Falcon H1R 7B Advanced reasoning in mathematics and coding; combines attention-based and state-space architectures for efficient sequence modeling under large batch sizes and long response lengths
Falcon-H1-Arabic High-performance Arabic AI, described as "a complete departure from traditional large language models"
Falcon-H1 Hybrid architecture based on Transformer and Mamba, aimed at removing the trade-off between performance and efficiency
Falcon Arabic Language AI for Modern Standard Arabic and dialects

A separate "Latest Updates" block announces newer releases: Falcon-Emirati (a 7-billion-parameter model built for Emirati vocabulary, expressions, and cultural context), Falcon-ASR (Emirati and Arabic speech transcription), and Falcon-OCR-Arabic (text extraction from images via a lightweight model).

A practical way to explore

  1. Scan the model cards first. Each card gives you the model's purpose in one or two lines, which is enough to rule models in or out before you click.
  2. Open the "Learn More" page for any model that fits your task. For example, if your task is extracting text from Arabic images, the Falcon-OCR-Arabic announcement points you toward that model; if it is image understanding from prompts, start with Falcon Perception.
  3. Check Latest Updates and Insights before deciding. Newer models are announced there, and a recent release may overlap with an older card. The Emirati, ASR, and OCR-Arabic models appear in updates rather than the main card grid.
  4. Confirm access details on the model page. The homepage does not state pricing, licensing terms, or how to obtain weights. Those details, if present, sit on the individual model pages.

Common sticking points

  • Expecting a download button on the homepage. The site is structured around model insight pages; the homepage routes you to them rather than hosting files.
  • Missing newer models. If you only read the main card grid, you can miss the Emirati, ASR, and OCR-Arabic releases, which are announced in the updates section.
  • Assuming cost or access. Nothing in the available page content confirms free access or open downloads, so verify on each model page before planning around it.

Choosing between models by task

  • Image and document understanding in natural language: Falcon Perception.
  • Math and coding reasoning in a compact model: Falcon H1R 7B.
  • Arabic text generation, including dialects: Falcon Arabic or Falcon-H1-Arabic.
  • Emirati-specific language and cultural context: Falcon-Emirati.
  • Arabic speech transcription: Falcon-ASR.
  • Text extraction from Arabic images: Falcon-OCR-Arabic.

These groupings follow the site's own descriptions; for benchmarks, context limits, or licensing, the individual model pages are the place to check.