Falcon 3
Technology Innovation InstituteTII's open-weights model line from Abu Dhabi. The original Falcon 3 family (1B/3B/7B/10B Apache-2.0, December 2024) has been extended in 2025-2026 by Falcon-H1 (a hybrid Mamba-Transformer architecture spanning 0.5B-34B with native 18-language support) and Falcon-H1 Arabic (3B/7B/34B, January 2026), positioned as the world's leading Arabic open model. Falcon H1R 7B (January 2026) outperforms Qwen and Nemotron variants up to 7x its size on reasoning. Adoption remains modest vs Llama/Qwen at equivalent size (Falcon3-7B-Instruct ~23K monthly HF downloads).
Falcon 3 family (1B, 3B, 7B, 10B), released 17 Dec 2024 by TII. Trained on 14T tokens. Sub-13B SLMs; now a dated generation relative to 2026 frontier.
Openness
3 high confidence- weights
- open
- data
- closed
- code
- closed
- license
- TII-Falcon-License-2.0(Apache-based+AUP,non-OSI)
Weights downloadable and freely usable commercially at any scale, corpus not released, no training code, so 3/open_weights. Corrected from 2/restricted on 2026-07-29 under issue #117: TII-Falcon-License-2.0 is Apache-based with an acceptable-use policy, and a conduct restriction caps neither who may use the model nor at what scale. It therefore sits in permissive_non_osi beside attribution requirements rather than in the capped tier beside Llama's 700M-MAU ceiling. Not 4 or 5: the corpus and the pipeline are both closed, and the license is non-OSI regardless, since the OSD forbids discriminating against a field of endeavor. The Hub reports this license as `other`, an abstention rather than an answer, so the tier rests on TII's published text.
- https://huggingface.co/api/models/tiiuae/Falcon3-10B-Base recorded 2026-07-29
ungated, 41.2GB of weight files, 7,688 downloads last month; cardData.license is `other`, so the Hub does not name the license and the tier rests on the TII text
- https://huggingface.co/blog/falcon3 recorded 2026-06-04
flagship phase-C verification source
- https://huggingface.co/tiiuae/Falcon3-7B-Base recorded 2026-06-04
flagship phase-C verification source
- https://www.tii.ae/news/falcon-3-uaes-technology-innovation-institute-launches-worlds-most-powerful-small-ai-models recorded 2026-06-04
flagship phase-C verification source
Adoption
2 high confidenceAvailable on Hugging Face with moderate uptake at launch as a sub-13B SLM; no large verified usage figures and relevance has faded against newer SLMs. Level 2 reflects modest reported traction.
- https://venturebeat.com/ai/uaes-falcon-3-challenges-open-source-leaders-amid-surging-demand-for-small-ai-models recorded 2026-06-04
flagship phase-C verification source
- https://huggingface.co/tiiuae/Falcon3-7B-Base recorded 2026-06-04
flagship phase-C verification source
Capability
2 high confidenceFalcon3-10B-Base: MMLU 73.1, MMLU-Pro 42.5, GSM8K 83.0; Falcon3-7B-Base: MMLU 67.4, MMLU-Pro 39.2. Competitive among sub-13B in 2024 but low vs 2026 frontier.
- https://huggingface.co/blog/falcon3 recorded 2026-06-04
flagship phase-C verification source
Unchanged since 2026-08-01 (last edited, not re-checked)