multimodal LLM
MERaLiON-2-3B
MERaLiON-2-3B is the lightweight member of the A*STAR MERaLiON-2 speech-text LLM family, balancing performance with lower compute.
Levelmodel
Classmultimodal LLM
Parameters3.47B
Contextnot in the notes
Released25 May 2025
Opennessopen weights restricted
Runs onnot in the notes
Statusactive
Benchmark Results
vendor-reported
AudioBenchspeech
Indonesian ASR, average WER by languagevendor-reported
0.132
00.080.16
Score in WER. Lower is better.
Where this comes from
Every score names who ran it: vendor-reported, maintainer-run or independent. A rank is shown with the population it was ranked in (gap rule 3).
Lineage
finetunedistillsuccessor
Gemma2-2b-IT is the parent (finetune)medium confidencemodel card
Whisper large-v3 is the parent (finetune)medium confidencemodel cardWhere this comes from
Parent and child relations from model cards, Hugging Face base_model fields and the genealogy tables of the research notes. The tree shows two generations each way, 6 models. Confidence and evidence are the notes' own.
R&D Topics
Runtimes
Identity
- Developer
- A*STAR (Institute for Infocomm Research, I2R)
- Family
- MERaLiON-2
- Released
- 25 May 2025 sourcesregistry date
- Weights published
- 25 May 2025
- Record checked
- 11 Oct 2026
Where this comes from
Developer and dates from the research notes. Release date is the vendor's public announcement; weights date is when open weights were published.
Architecture and Size
- Architecture
- other Localised Whisper-large-v3 speech encoder with a Gemma2-2b-IT text decoder
- Total parameters
- 3.47B sources
- Active parameters
- –
- Checkpoint parameters
- –
- Note
- Hugging Face API Safetensors count
- Context length
- not in the notes
- Input
- audio, text
- Output
- text
Where this comes from
Parameter counts as the developer states them. Where the checkpoint holds a different total, both are shown (register A16).
Openness and Access
- Openness
- open weights restricted
- Licence
- MERaLiON Public License v3 sources
- Access routes
- open weights download
Where this comes from
The licence file governs over registry tags and aggregator labels.
Deployment
- Runs on
- not in the notes
- Footprint
- not in the notes
- Runtimes
- vLLM
Where this comes from
Deployment tiers and runtimes as named in the notes. Footprint is the notes' figure, not a measurement by AURION.
Indonesia
- Indonesian support
- measured on a benchmark
- MERaLiON ASR table, Indonesian row (average word error rate by language): 0.131950, run by A*STAR MERaLiON team (developer-reported) huggingface.co
- Regional languages
- not in the notes
- Jakarta hosting
- unknown
- Languages
- Primarily English (global and Singapore) and Chinese, with audio support for Malay, Tamil, Indonesian, Thai and Vietnamese
Where this comes from
Indonesia fields are carried on every model; unknown is an allowed value.
Industry Use Cases
No industry use case links this model.
Links
ModelMERaLiON-2-3BA*STAR (Institute for Infocomm Research, I2R), multimodal LLM
MERaLiON-2-3BA*STAR (Institute for Infocomm Research, I2R), multimodal LLMR&D TopicsLow-Resource and Regional Language ResourcesRegional or National Foundation Model Training (Continued Pretraining at 7B to 70B)Audio Language Models for Speech Understanding
Use CasesIndonesian and Regional-Language LLMsPublic SectorIndonesian NLU (Classification, NER)Public SectorBanking Customer Service AssistantsFinanceCitizen Service AssistantsPublic Sector+5 moreUse cases of models in the same R&D topic; none is recorded for this model itself.
Who Runs It
PT Bank Tabungan Negara (BTN)In production, 1 Jul 2026
PT PegadaianIn production, 5 Jun 2026
Bank Central Asia (BCA)In production, 16 Mar 2026Bank Central Asia (BCA)In production, 16 Mar 2026+13 moreDeployments of these use cases; none names this model.
PT Bank Tabungan Negara (BTN)In production, 1 Jul 2026PT PegadaianIn production, 5 Jun 2026Bank Central Asia (BCA)In production, 16 Mar 2026Bank Central Asia (BCA)In production, 16 Mar 2026+13 moreDeployments of these use cases; none names this model.Models 3
R&D topics 3
Regional or National Foundation Model Training (Continued Pretraining at 7B to 70B)Pretraining or continued pretraining a mid-size model to serve a national or regional lan…produced byLow-Resource and Regional Language ResourcesDatasets, benchmarks and adaptation for under-served languages.produced byAudio Language Models for Speech UnderstandingLanguage models that take audio as input to transcribe, answer questions about and conver…produced by
Sources
- huggingface.co/MERaLiON/MERaLiON-2-3B https://huggingface.co/MERaLiON/MERaLiON-2-3B
- huggingface.co/MERaLiON/MERaLiON-2-10B https://huggingface.co/MERaLiON/MERaLiON-2-10B
- huggingface.co/api/models/MERaLiON/MERaLiON-2-3B https://huggingface.co/api/models/MERaLiON/MERaLiON-2-3B
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "MERaLiON-2-3B", https://aurion.semai.id/models/a-star-meralion-2-3b/, records of October 2026.