Skip to main content
Rase AI Intelligence Newsletter

Models & assistants Model or AI service

Falcon-H1 combines attention and state-space layers

The Technology Innovation Institute built Falcon-H1 as a family of downloadable language models with a hybrid architecture. The family includes multiple sizes and offers base and instruction-tuned checkpoints. [2][3]

Model access
Multiple base and instruction-tuned checkpoints are downloadable, including the 34B Instruct version. [2][3]
License
Falcon-LLM License for the documented 34B Instruct weights. [3]

What the model does

TII’s release repository dates the Falcon-H1 family release to May 21, 2025. It lists variants from 0.5B through 34B parameters and documents which checkpoints support longer text inputs. [2][3]

Why it matters

Combining two approaches to processing text gives teams another architecture to test for long documents and multi-turn conversations. Model size and supported context both affect the infrastructure required. [2][3]

Who can use it

Researchers and developers can choose a base model for further training or an instruction-tuned checkpoint for direct evaluation. The 34B model card describes one option in the family. [2][3]

Key dates

  1. Falcon-H1 family released [1][2]

Ways to use it

For a long-document task, compare answer quality and memory use at several input lengths on your chosen checkpoint. Select the specific size and instruction tuning suited to the task. [2][3]

Model limitations

The highest context limit applies only to supported variants, and actual memory use depends on the serving setup. TII’s Falcon-LLM License applies to the 34B Instruct checkpoint. [2][3]

Sources and verification

Catalog evidence checked How we verify records

  1. Primary source ATRC’s May 21 Falcon-H1 and Falcon Arabic unveiling
  2. Primary source TII’s Falcon-H1 versioned model release and checkpoint listing
  3. Primary source tiiuae/Falcon-H1-34B-Instruct
  4. Other source Falcon-H1: A Family of Hybrid-Head Language Models Redefining Efficiency and Performance
  5. Primary source Hugging Face registry metadata for tiiuae/Falcon-H1-34B-Instruct

Related profiles

Models & assistants

GPT-6.2 Astra and Sol

OpenAI

Astra targets demanding projects. Sol targets everyday tasks. Both support text and image input in the API.

High confidenceEvidence checked

Models & assistants

MAI Code 2 and Code 2 Flash

Microsoft AI

Code 2 targets complex software work. Flash targets frequent edits, explanations and routine fixes.

High confidenceEvidence checked

Models & assistants

MAI Realtime 1

Microsoft AI

Users can try a MAI Labs demo. Developers can register for a limited Foundry preview.

High confidenceEvidence checked