Language modelLimited access
Falcon-Emirati
Technology Innovation Institute
- Released
- October 6, 2026
- Data date
- October 7, 2026
- Input
- Text
- Output
- Text
Falcon-Emirati focuses language processing on Emirati Arabic, including regional vocabulary, idioms, and cultural references. TII names Falcon-H1-Arabic as its base and provides access through Falcon Chat.
TII reports 84.83% on the Emirati Alyah benchmark for the 7B model. The provider released it on October 6, 2026 alongside its new Arabic speech-recognition and document-processing models.
Specifications and access
| Specification | Value and source |
|---|---|
| Model class | Language model for Emirati Arabic, including regional vocabulary, idioms, and cultural contextSource |
| Input modalities | TextSource |
| Output modalities | TextSource |
| Base model | Falcon-H1-ArabicSource |
| Access | TII names Falcon Chat as the access route; the announcement does not list a direct model API endpointSource |
Page 1 of 3
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
Alyah (Show measurement, test conditions, and source)
- Source value
- 84.83
- Score
- 84.83
- Metric
- native Emirati Arabic multiple-choice accuracy
- Unit
- %
- Category
- provider-reported-benchmarks
- Direction
- Higher is better
- Sample
- 1173
- Source type
- vendor-reported
- Evaluator
- Technology Innovation Institute Falcon-Emirati page
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Technology Innovation Institute report for exact checkpoint Falcon-Emirati-7B. The TII page specifies a 1,173-item native Emirati Arabic multiple-choice benchmark.
- Context
- Provider-reported result; no independent evaluation configuration or confidence interval is given on the page.
Emirati Arabic dialect fidelity (Show measurement, test conditions, and source)
- Source value
- 52.1
- Score
- 52.1
- Metric
- LLM-judged naturalness / dialect-fidelity score with partial credit
- Unit
- %
- Category
- provider-reported-benchmarks
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Technology Innovation Institute Falcon-Emirati page
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Technology Innovation Institute report for exact checkpoint Falcon-Emirati-7B. The TII page identifies this as an LLM-judged partial-credit score; it does not publish a sample count or judge protocol.
- Context
- Provider-reported result; no independent evaluation configuration or confidence interval is given on the page.
ArabCulture-Dialogue · UAE subset (Show measurement, test conditions, and source)
- Source value
- 85.57
- Score
- 85.57
- Metric
- cultural-understanding score
- Unit
- %
- Category
- provider-reported-benchmarks
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Technology Innovation Institute Falcon-Emirati page
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Technology Innovation Institute report for exact checkpoint Falcon-Emirati-7B. The TII page identifies the UAE subset of ArabCulture-Dialogue but does not publish a sample count or scoring protocol.
- Context
- Provider-reported result; no independent evaluation configuration or confidence interval is given on the page.
LLM-judged correctness (Show measurement, test conditions, and source)
- Source value
- 48.8
- Score
- 48.8
- Metric
- LLM-judged correctness score
- Unit
- %
- Category
- provider-reported-benchmarks
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Technology Innovation Institute Falcon-Emirati page
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Technology Innovation Institute report for exact checkpoint Falcon-Emirati-7B. The TII comparison table reports this metric but does not publish a sample count or judge protocol.
- Context
- Provider-reported result; no independent evaluation configuration or confidence interval is given on the page.
Alyah · open-ended correctness · strict (Show measurement, test conditions, and source)
- Source value
- 0.42
- Score
- 0.42
- Metric
- LLM-judged correctness, strict pass/fail
- Unit
- ratio
- Benchmark version
- Alyah open-ended evaluation, Gemini 3.7 Flash
- Category
- provider-reported-benchmarks
- Direction
- Higher is better
- Sample
- 1173
- Source type
- vendor-reported
- Evaluator
- Technology Innovation Institute Falcon-Emirati evaluation article (Hugging Face)
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. TII reports open-ended generations for the exact Falcon-Emirati-7B checkpoint, judged with Gemini 3.7 Flash on the same 1,173 Alyah questions. This is the strict pass/fail correctness score, separate from the partial-credit score.
- Context
- Provider-reported LLM-judge result, not an independent human evaluation. The chart rounds to two decimals; no confidence interval is given.
Alyah · dialect fidelity · Religious & Social Sensitivity (Show measurement, test conditions, and source)
- Source value
- 0.49
- Score
- 0.49
- Metric
- LLM-judged Emirati-dialect fidelity, partial credit
- Unit
- ratio
- Benchmark version
- Alyah open-ended evaluation, Gemini 3.7 Flash
- Category
- provider-reported-benchmarks
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Technology Innovation Institute Falcon-Emirati evaluation article (Hugging Face)
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. TII reports open-ended generations for the exact Falcon-Emirati-7B checkpoint, judged with Gemini 3.7 Flash on the same 1,173 Alyah questions. This is the partial-credit dialect-fidelity score for Religious & Social Sensitivity; the chart does not publish a sample count for this category.
- Context
- TII reports this vendor-run judge score rounded to two decimal places. It is not gold-label accuracy; the category sample count and confidence interval are not published.
Alyah · dialect fidelity · Poetry & Creative Expression (Show measurement, test conditions, and source)
- Source value
- 0.59
- Score
- 0.59
- Metric
- LLM-judged Emirati-dialect fidelity, partial credit
- Unit
- ratio
- Benchmark version
- Alyah open-ended evaluation, Gemini 3.7 Flash
- Category
- provider-reported-benchmarks
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Technology Innovation Institute Falcon-Emirati evaluation article (Hugging Face)
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. TII reports open-ended generations for the exact Falcon-Emirati-7B checkpoint, judged with Gemini 3.7 Flash on the same 1,173 Alyah questions. This is the partial-credit dialect-fidelity score for Poetry & Creative Expression; the chart does not publish a sample count for this category.
- Context
- TII reports this vendor-run judge score rounded to two decimal places. It is not gold-label accuracy; the category sample count and confidence interval are not published.
Alyah · dialect fidelity · Language & Dialect (Show measurement, test conditions, and source)
- Source value
- 0.51
- Score
- 0.51
- Metric
- LLM-judged Emirati-dialect fidelity, partial credit
- Unit
- ratio
- Benchmark version
- Alyah open-ended evaluation, Gemini 3.7 Flash
- Category
- provider-reported-benchmarks
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Technology Innovation Institute Falcon-Emirati evaluation article (Hugging Face)
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. TII reports open-ended generations for the exact Falcon-Emirati-7B checkpoint, judged with Gemini 3.7 Flash on the same 1,173 Alyah questions. This is the partial-credit dialect-fidelity score for Language & Dialect; the chart does not publish a sample count for this category.
- Context
- TII reports this vendor-run judge score rounded to two decimal places. It is not gold-label accuracy; the category sample count and confidence interval are not published.
Alyah · dialect fidelity · Imagery & Figurative Meaning (Show measurement, test conditions, and source)
- Source value
- 0.55
- Score
- 0.55
- Metric
- LLM-judged Emirati-dialect fidelity, partial credit
- Unit
- ratio
- Benchmark version
- Alyah open-ended evaluation, Gemini 3.7 Flash
- Category
- provider-reported-benchmarks
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Technology Innovation Institute Falcon-Emirati evaluation article (Hugging Face)
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. TII reports open-ended generations for the exact Falcon-Emirati-7B checkpoint, judged with Gemini 3.7 Flash on the same 1,173 Alyah questions. This is the partial-credit dialect-fidelity score for Imagery & Figurative Meaning; the chart does not publish a sample count for this category.
- Context
- TII reports this vendor-run judge score rounded to two decimal places. It is not gold-label accuracy; the category sample count and confidence interval are not published.
Alyah · dialect fidelity · Historical & Heritage Knowledge (Show measurement, test conditions, and source)
- Source value
- 0.48
- Score
- 0.48
- Metric
- LLM-judged Emirati-dialect fidelity, partial credit
- Unit
- ratio
- Benchmark version
- Alyah open-ended evaluation, Gemini 3.7 Flash
- Category
- provider-reported-benchmarks
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Technology Innovation Institute Falcon-Emirati evaluation article (Hugging Face)
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. TII reports open-ended generations for the exact Falcon-Emirati-7B checkpoint, judged with Gemini 3.7 Flash on the same 1,173 Alyah questions. This is the partial-credit dialect-fidelity score for Historical & Heritage Knowledge; the chart does not publish a sample count for this category.
- Context
- TII reports this vendor-run judge score rounded to two decimal places. It is not gold-label accuracy; the category sample count and confidence interval are not published.
Alyah · dialect fidelity · Greetings & Daily Expressions (Show measurement, test conditions, and source)
- Source value
- 0.47
- Score
- 0.47
- Metric
- LLM-judged Emirati-dialect fidelity, partial credit
- Unit
- ratio
- Benchmark version
- Alyah open-ended evaluation, Gemini 3.7 Flash
- Category
- provider-reported-benchmarks
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Technology Innovation Institute Falcon-Emirati evaluation article (Hugging Face)
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. TII reports open-ended generations for the exact Falcon-Emirati-7B checkpoint, judged with Gemini 3.7 Flash on the same 1,173 Alyah questions. This is the partial-credit dialect-fidelity score for Greetings & Daily Expressions; the chart does not publish a sample count for this category.
- Context
- TII reports this vendor-run judge score rounded to two decimal places. It is not gold-label accuracy; the category sample count and confidence interval are not published.
Alyah · dialect fidelity · Etiquette & Values (Show measurement, test conditions, and source)
- Source value
- 0.58
- Score
- 0.58
- Metric
- LLM-judged Emirati-dialect fidelity, partial credit
- Unit
- ratio
- Benchmark version
- Alyah open-ended evaluation, Gemini 3.7 Flash
- Category
- provider-reported-benchmarks
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Technology Innovation Institute Falcon-Emirati evaluation article (Hugging Face)
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. TII reports open-ended generations for the exact Falcon-Emirati-7B checkpoint, judged with Gemini 3.7 Flash on the same 1,173 Alyah questions. This is the partial-credit dialect-fidelity score for Etiquette & Values; the chart does not publish a sample count for this category.
- Context
- TII reports this vendor-run judge score rounded to two decimal places. It is not gold-label accuracy; the category sample count and confidence interval are not published.
Page 1 of 3
Sources and data date
| Type | Evidence and data date |
|---|---|
| Input | Documented modalities. Retrieved October 7, 2026. |
| Output | Documented modalities. Retrieved October 7, 2026. |
| Research status | Research date October 7, 2026. 2 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not found in the inspected sources | independent-model-evaluation Original wording The official TII product page and TII-authored Hugging Face evaluation article were read. They add provider-reported results, including blind Gemini 3.7 Flash comparisons, but no independent evaluator score for the exact Falcon-Emirati-7B checkpoint was found. This records the search outcome, not proof that no independent evaluation exists. |
| Additional source | TII Falcon model launch (retrieved October 7, 2026) · TII Falcon model launch · Editorial description reviewed October 7, 2026 |
| Additional source | Falcon Chat (retrieved October 7, 2026) |
| Additional source | Technology Innovation Institute Falcon-Emirati evaluation article (Hugging Face) (retrieved October 7, 2026) |