AI SingaporeLanguage model
Gemma-SEA-LION-v4-27B-IT
Gemma-SEA-LION-v4-27B-IT has no overall index value yet because there are too few independent benchmark results.
PopularityRank 535 of 785
30out of 100+1 vs. last month
based on search, usage and attention
Summary
Gemma-SEA-LION-v4-27B-IT specializes Gemma 3 27B for ten Southeast Asian languages. AI Singapore describes roughly 10 million post-training question and answer examples in Burmese, English, Indonesian, Khmer, Lao, Malay, Tagalog, Tamil, Thai, and Vietnamese. Its 128K context and Gemma license target regional self-hosted applications.
The current model card also describes inherited image and document understanding plus function calling and structured outputs. The card also warns that the model is not fully safety aligned and needs application-level safeguards.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 36 published benchmark results
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.
SEA-SafeguardBench prompt AUPRC, General English (Show measurement, test conditions, and source)
- Source value
- 90.9
- Score
- 90.9
- Metric
- AUPRC, reported on the paper's percent scale
- Unit
- %
- Benchmark version
- SEA-SafeguardBench, ACL 2026 Findings paper, Table 2
- Category
- safety
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- SEA-SafeguardBench ACL 2026 Findings paper (AI Singapore co-author)
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. The paper's Appendix B.3 names the zero-shot checkpoint as Gemma-SEA-LION-v4-it 27B. The pinned official card identifies the matching instruction-tuned checkpoint as aisingapore/Gemma-SEA-LION-v4-27B-IT; the paper does not state the immutable weight revision used. This is prompt classification on the General subset, English. AUPRC is calculated from confidence scores for representative tokens.
- Context
- The study is peer-reviewed, but an AI Singapore-affiliated coauthor is listed and AI Singapore developed this checkpoint; this is therefore classified as vendor-reported, not an independent evaluation. This is a safety-classification result, not a measure of general language quality. The paper maps the sensitive label to safe for prompt classification and supplies no model-specific sample count, run count or confidence interval. It does not pin the exact weight revision used.
SEA-SafeguardBench prompt AUPRC, General SEA (Show measurement, test conditions, and source)
- Source value
- 88.5
- Score
- 88.5
- Metric
- AUPRC, reported on the paper's percent scale
- Unit
- %
- Benchmark version
- SEA-SafeguardBench, ACL 2026 Findings paper, Table 2
- Category
- safety
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- SEA-SafeguardBench ACL 2026 Findings paper (AI Singapore co-author)
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. The paper's Appendix B.3 names the zero-shot checkpoint as Gemma-SEA-LION-v4-it 27B. The pinned official card identifies the matching instruction-tuned checkpoint as aisingapore/Gemma-SEA-LION-v4-27B-IT; the paper does not state the immutable weight revision used. This is prompt classification on the General subset, SEA. AUPRC is calculated from confidence scores for representative tokens.
- Context
- The study is peer-reviewed, but an AI Singapore-affiliated coauthor is listed and AI Singapore developed this checkpoint; this is therefore classified as vendor-reported, not an independent evaluation. This is a safety-classification result, not a measure of general language quality. The paper maps the sensitive label to safe for prompt classification and supplies no model-specific sample count, run count or confidence interval. It does not pin the exact weight revision used.
SEA-SafeguardBench prompt AUPRC, ITW Cultural English (Show measurement, test conditions, and source)
- Source value
- 98.2
- Score
- 98.2
- Metric
- AUPRC, reported on the paper's percent scale
- Unit
- %
- Benchmark version
- SEA-SafeguardBench, ACL 2026 Findings paper, Table 2
- Category
- safety
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- SEA-SafeguardBench ACL 2026 Findings paper (AI Singapore co-author)
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. The paper's Appendix B.3 names the zero-shot checkpoint as Gemma-SEA-LION-v4-it 27B. The pinned official card identifies the matching instruction-tuned checkpoint as aisingapore/Gemma-SEA-LION-v4-27B-IT; the paper does not state the immutable weight revision used. This is prompt classification on the In-the-Wild Cultural subset, English. AUPRC is calculated from confidence scores for representative tokens.
- Context
- The study is peer-reviewed, but an AI Singapore-affiliated coauthor is listed and AI Singapore developed this checkpoint; this is therefore classified as vendor-reported, not an independent evaluation. This is a safety-classification result, not a measure of general language quality. The paper maps the sensitive label to safe for prompt classification and supplies no model-specific sample count, run count or confidence interval. It does not pin the exact weight revision used.
SEA-SafeguardBench prompt AUPRC, ITW Cultural SEA (Show measurement, test conditions, and source)
- Source value
- 97.4
- Score
- 97.4
- Metric
- AUPRC, reported on the paper's percent scale
- Unit
- %
- Benchmark version
- SEA-SafeguardBench, ACL 2026 Findings paper, Table 2
- Category
- safety
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- SEA-SafeguardBench ACL 2026 Findings paper (AI Singapore co-author)
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. The paper's Appendix B.3 names the zero-shot checkpoint as Gemma-SEA-LION-v4-it 27B. The pinned official card identifies the matching instruction-tuned checkpoint as aisingapore/Gemma-SEA-LION-v4-27B-IT; the paper does not state the immutable weight revision used. This is prompt classification on the In-the-Wild Cultural subset, SEA. AUPRC is calculated from confidence scores for representative tokens.
- Context
- The study is peer-reviewed, but an AI Singapore-affiliated coauthor is listed and AI Singapore developed this checkpoint; this is therefore classified as vendor-reported, not an independent evaluation. This is a safety-classification result, not a measure of general language quality. The paper maps the sensitive label to safe for prompt classification and supplies no model-specific sample count, run count or confidence interval. It does not pin the exact weight revision used.
SEA-SafeguardBench prompt AUPRC, CG Cultural English (Show measurement, test conditions, and source)
- Source value
- 65.4
- Score
- 65.4
- Metric
- AUPRC, reported on the paper's percent scale
- Unit
- %
- Benchmark version
- SEA-SafeguardBench, ACL 2026 Findings paper, Table 2
- Category
- safety
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- SEA-SafeguardBench ACL 2026 Findings paper (AI Singapore co-author)
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. The paper's Appendix B.3 names the zero-shot checkpoint as Gemma-SEA-LION-v4-it 27B. The pinned official card identifies the matching instruction-tuned checkpoint as aisingapore/Gemma-SEA-LION-v4-27B-IT; the paper does not state the immutable weight revision used. This is prompt classification on the Content Generation Cultural subset, English. AUPRC is calculated from confidence scores for representative tokens.
- Context
- The study is peer-reviewed, but an AI Singapore-affiliated coauthor is listed and AI Singapore developed this checkpoint; this is therefore classified as vendor-reported, not an independent evaluation. This is a safety-classification result, not a measure of general language quality. The paper maps the sensitive label to safe for prompt classification and supplies no model-specific sample count, run count or confidence interval. It does not pin the exact weight revision used.
SEA-SafeguardBench prompt AUPRC, CG Cultural SEA (Show measurement, test conditions, and source)
- Source value
- 64.7
- Score
- 64.7
- Metric
- AUPRC, reported on the paper's percent scale
- Unit
- %
- Benchmark version
- SEA-SafeguardBench, ACL 2026 Findings paper, Table 2
- Category
- safety
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- SEA-SafeguardBench ACL 2026 Findings paper (AI Singapore co-author)
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. The paper's Appendix B.3 names the zero-shot checkpoint as Gemma-SEA-LION-v4-it 27B. The pinned official card identifies the matching instruction-tuned checkpoint as aisingapore/Gemma-SEA-LION-v4-27B-IT; the paper does not state the immutable weight revision used. This is prompt classification on the Content Generation Cultural subset, SEA. AUPRC is calculated from confidence scores for representative tokens.
- Context
- The study is peer-reviewed, but an AI Singapore-affiliated coauthor is listed and AI Singapore developed this checkpoint; this is therefore classified as vendor-reported, not an independent evaluation. This is a safety-classification result, not a measure of general language quality. The paper maps the sensitive label to safe for prompt classification and supplies no model-specific sample count, run count or confidence interval. It does not pin the exact weight revision used.
SEA-SafeguardBench prompt AUPRC, average English (Show measurement, test conditions, and source)
- Source value
- 84.2
- Score
- 84.2
- Metric
- AUPRC, reported on the paper's percent scale
- Unit
- %
- Benchmark version
- SEA-SafeguardBench, ACL 2026 Findings paper, Table 2
- Category
- safety
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- SEA-SafeguardBench ACL 2026 Findings paper (AI Singapore co-author)
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. The paper's Appendix B.3 names the zero-shot checkpoint as Gemma-SEA-LION-v4-it 27B. The pinned official card identifies the matching instruction-tuned checkpoint as aisingapore/Gemma-SEA-LION-v4-27B-IT; the paper does not state the immutable weight revision used. This is the prompt-classification average for English in Table 2. AUPRC is calculated from confidence scores for representative tokens.
- Context
- The study is peer-reviewed, but an AI Singapore-affiliated coauthor is listed and AI Singapore developed this checkpoint; this is therefore classified as vendor-reported, not an independent evaluation. This is a safety-classification result, not a measure of general language quality. The paper maps the sensitive label to safe for prompt classification and supplies no model-specific sample count, run count or confidence interval. It does not pin the exact weight revision used.
SEA-SafeguardBench prompt AUPRC, average SEA (Show measurement, test conditions, and source)
- Source value
- 85
- Score
- 85
- Metric
- AUPRC, reported on the paper's percent scale
- Unit
- %
- Benchmark version
- SEA-SafeguardBench, ACL 2026 Findings paper, Table 2
- Category
- safety
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- SEA-SafeguardBench ACL 2026 Findings paper (AI Singapore co-author)
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. The paper's Appendix B.3 names the zero-shot checkpoint as Gemma-SEA-LION-v4-it 27B. The pinned official card identifies the matching instruction-tuned checkpoint as aisingapore/Gemma-SEA-LION-v4-27B-IT; the paper does not state the immutable weight revision used. This is the prompt-classification average for SEA in Table 2. AUPRC is calculated from confidence scores for representative tokens.
- Context
- The study is peer-reviewed, but an AI Singapore-affiliated coauthor is listed and AI Singapore developed this checkpoint; this is therefore classified as vendor-reported, not an independent evaluation. This is a safety-classification result, not a measure of general language quality. The paper maps the sensitive label to safe for prompt classification and supplies no model-specific sample count, run count or confidence interval. It does not pin the exact weight revision used.
SEA-SafeguardBench response AUPRC, General English (Show measurement, test conditions, and source)
- Source value
- 85.2
- Score
- 85.2
- Metric
- AUPRC, reported on the paper's percent scale
- Unit
- %
- Benchmark version
- SEA-SafeguardBench, ACL 2026 Findings paper, Table 2
- Category
- safety
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- SEA-SafeguardBench ACL 2026 Findings paper (AI Singapore co-author)
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. The paper's Appendix B.3 names the zero-shot checkpoint as Gemma-SEA-LION-v4-it 27B. The pinned official card identifies the matching instruction-tuned checkpoint as aisingapore/Gemma-SEA-LION-v4-27B-IT; the paper does not state the immutable weight revision used. This is response classification on the General subset, English. AUPRC is calculated from confidence scores for representative tokens.
- Context
- The study is peer-reviewed, but an AI Singapore-affiliated coauthor is listed and AI Singapore developed this checkpoint; this is therefore classified as vendor-reported, not an independent evaluation. This is a safety-classification result, not a measure of general language quality. The paper maps the sensitive label to harmful for response classification and supplies no model-specific sample count, run count or confidence interval. It does not pin the exact weight revision used.
SEA-SafeguardBench response AUPRC, General SEA (Show measurement, test conditions, and source)
- Source value
- 68.7
- Score
- 68.7
- Metric
- AUPRC, reported on the paper's percent scale
- Unit
- %
- Benchmark version
- SEA-SafeguardBench, ACL 2026 Findings paper, Table 2
- Category
- safety
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- SEA-SafeguardBench ACL 2026 Findings paper (AI Singapore co-author)
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. The paper's Appendix B.3 names the zero-shot checkpoint as Gemma-SEA-LION-v4-it 27B. The pinned official card identifies the matching instruction-tuned checkpoint as aisingapore/Gemma-SEA-LION-v4-27B-IT; the paper does not state the immutable weight revision used. This is response classification on the General subset, SEA. AUPRC is calculated from confidence scores for representative tokens.
- Context
- The study is peer-reviewed, but an AI Singapore-affiliated coauthor is listed and AI Singapore developed this checkpoint; this is therefore classified as vendor-reported, not an independent evaluation. This is a safety-classification result, not a measure of general language quality. The paper maps the sensitive label to harmful for response classification and supplies no model-specific sample count, run count or confidence interval. It does not pin the exact weight revision used.
SEA-SafeguardBench response AUPRC, CG Cultural English (Show measurement, test conditions, and source)
- Source value
- 63.8
- Score
- 63.8
- Metric
- AUPRC, reported on the paper's percent scale
- Unit
- %
- Benchmark version
- SEA-SafeguardBench, ACL 2026 Findings paper, Table 2
- Category
- safety
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- SEA-SafeguardBench ACL 2026 Findings paper (AI Singapore co-author)
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. The paper's Appendix B.3 names the zero-shot checkpoint as Gemma-SEA-LION-v4-it 27B. The pinned official card identifies the matching instruction-tuned checkpoint as aisingapore/Gemma-SEA-LION-v4-27B-IT; the paper does not state the immutable weight revision used. This is response classification on the Content Generation Cultural subset, English. AUPRC is calculated from confidence scores for representative tokens.
- Context
- The study is peer-reviewed, but an AI Singapore-affiliated coauthor is listed and AI Singapore developed this checkpoint; this is therefore classified as vendor-reported, not an independent evaluation. This is a safety-classification result, not a measure of general language quality. The paper maps the sensitive label to harmful for response classification and supplies no model-specific sample count, run count or confidence interval. It does not pin the exact weight revision used.
SEA-SafeguardBench response AUPRC, CG Cultural SEA (Show measurement, test conditions, and source)
- Source value
- 75.7
- Score
- 75.7
- Metric
- AUPRC, reported on the paper's percent scale
- Unit
- %
- Benchmark version
- SEA-SafeguardBench, ACL 2026 Findings paper, Table 2
- Category
- safety
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- SEA-SafeguardBench ACL 2026 Findings paper (AI Singapore co-author)
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. The paper's Appendix B.3 names the zero-shot checkpoint as Gemma-SEA-LION-v4-it 27B. The pinned official card identifies the matching instruction-tuned checkpoint as aisingapore/Gemma-SEA-LION-v4-27B-IT; the paper does not state the immutable weight revision used. This is response classification on the Content Generation Cultural subset, SEA. AUPRC is calculated from confidence scores for representative tokens.
- Context
- The study is peer-reviewed, but an AI Singapore-affiliated coauthor is listed and AI Singapore developed this checkpoint; this is therefore classified as vendor-reported, not an independent evaluation. This is a safety-classification result, not a measure of general language quality. The paper maps the sensitive label to harmful for response classification and supplies no model-specific sample count, run count or confidence interval. It does not pin the exact weight revision used.
Page 1 of 3
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
The 100 most popular models of this category, 0 to 100, higher is more popular
20 of 101 models
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 7, 2026
Gemma-SEA-LION-v4-27B-ITYi-1.5 34B 30Kimi K2 0905 30GPT-5 Pro 30Baichuan M2 32B 30
Method changes in the timeline, a jump at these points is no market event:
- 1November 1, 2026: Value only with search or usage data, 30-day attention
What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.
Similar models
More models of the same kind as Gemma-SEA-LION-v4-27B-IT, chosen by model line, developer, size, price and release date.
- Qwen3 Omni 30B A3B ThinkingAlibaba · similar size
- Qwen 3 VL 30B A3B ThinkingAlibaba · similar size
- Qwen 3 VL 30B A3BAlibaba · similar size
- Magistral Small 1.2Mistral AI · similar size
- Qwen 3.5 27BAlibaba · similar size
- Granite-4.0-H-SmallIBM · similar size
- Trinity MiniArcee AI · similar size
- Qwen 3.6 27BAlibaba · similar size
All specifications
Architecture
Input and output
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
Frequently asked questions about Gemma-SEA-LION-v4-27B-IT
How large is the context window of Gemma-SEA-LION-v4-27B-IT?
128,000 tokens. That is how much text Gemma-SEA-LION-v4-27B-IT reads at once, about 190 A4 pages.
Page count as a rule of thumb from Artificial Analysis
GLM-5.1200K
DeepSeek-V3 0324164K
GLM-4.5131K
GPT OSS 120B131K
Gemma-SEA-LION-v4-27B-ITRank 98 of 103128K
DeepSeek-V3.2128K
DeepSeek-R1128K
Can I run Gemma-SEA-LION-v4-27B-IT locally?
Yes, the weights are open.
Stated by AI Singapore
- Open weights
- yes
- License
- Gemma Terms of Use
- Parameters
- 27B
- active per token
- n/a
Sources and data date
Show all 14 sources
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.
Without vendor claims and historical or disputed results; those stay labelled in the single values.
| Type | Evidence and data date |
|---|---|
| Input | Documented modalities. Retrieved October 27, 2026. |
| Output | Documented modalities. Retrieved October 27, 2026. |
| Research status | Research date October 15, 2026. 7 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not found in the inspected sources | current-exact-checkpoint-leaderboard-row Note The live SEA-HELM roster read on 2026-10-15 lists newer v4.5 variants but not Gemma-SEA-LION-v4-27B-IT. The dated card figure is retained; no v4.5 or other Gemma score is used as a substitute. |
| Not found in the inspected sources | current-exact-checkpoint-leaderboard-row Note The current SEA-HELM roster still has no row for Gemma-SEA-LION-v4-27B-IT. Its dated historical card figure remains in the earlier bundle; no v4.5 or sibling score is substituted. |
| Unresolved | paper-checkpoint-revision-and-evaluation-uncertainty Note The paper's model-name and configuration text support the instruction-tuned profile match, but it supplies no immutable weight revision, model-specific sample count, repeated-run count or confidence interval for this row. None is inferred from the overall dataset totals. |
| Additional source | Hugging Face (retrieved October 15, 2026) · AI Singapore Gemma SEA LION v4 model card · Editorial description reviewed October 2, 2026 |
| Additional source | AI Singapore Gemma-SEA-LION-v4-27B-IT card leaderboard figure (retrieved October 15, 2026) |
| Additional source | SEA-SafeguardBench ACL 2026 Findings paper (AI Singapore co-author) (retrieved October 16, 2026; October 17, 2026; October 18, 2026) |
| Additional source | SEA-HELM official leaderboard, embedded result data (retrieved October 15, 2026; October 16, 2026; October 17, 2026) |
| Additional source | Model metadata (retrieved October 16, 2026) |
| Additional source | Model metadata (retrieved October 17, 2026)12 |
| Additional source | OpenRouter: Popularity index (derived), CC BY 4.0 |
| Additional source | Vercel AI Gateway: Popularity index (derived), CC BY 4.0 |
| Additional source | Ollama: cumulative downloads for the popularity index |
| Additional source | Replicate: cumulative runs for the popularity index |
| Additional source | Google Trending Now: limited feed selection for Germany and the US |
| Additional source | Creative Commons Attribution 4.0 |