MaLLaM
Malaysia's open-weight Malay large language model, pre-trained from scratch on Malaysian text by Mesolitica.
Overview
MaLLaM (Malaysia Large Language Model) is a family of open-weight foundation models from Malaysian AI startup Mesolitica, pre-trained from scratch on a roughly 90-billion-token corpus of Malaysian text using the Mistral architecture. It ships in 1.1B, 3B and 5B parameter sizes and is tuned to understand Bahasa Malaysia plus local slang, dialects and regional languages. Mesolitica later trained a localized generative model on AWS using Trainium chips. Weights are released on Hugging Face.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 24 Jul 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
ASEAN Perspective
MaLLaM in Southeast Asia
ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).
What this is for: A sovereign Malay-language LLM that captures local slang, dialects and cultural context for AI assistants and text applications.
Who this is for: Malaysian developers, enterprises and public agencies needing Malay-first generative AI rather than an English model with a translation layer.
Availability: Free open-weights on Hugging Face; base models require further fine-tuning for downstream tasks.
What people say
You won't find MaLLaM on G2, Capterra, or Trustpilot. It's a set of open base models, not a SaaS product; its reputation is forged on Hugging Face and within the Malaysian AI community. Mesolitica ships several checkpoints (mallam-1.1B, 3B and 5B at 4096 context, plus instruction-tuned variants), and the models are actively maintained, with the 1.1B checkpoint updated as recently as 2026. Mesolitica also publishes a Malay LLM Leaderboard, which is the closest thing to a public scoreboard for this niche.
The credibility hook practitioners cite is that MaLLaM was pre-trained from scratch on a roughly 90-billion-token Malaysian corpus rather than fine-tuned over an English base — so its representations capture local slang and dialect rather than translating through English. For a sovereign-language effort from a small startup, that's a genuine engineering claim.
The caveats are real: these are small models (1–5B) that will feel weak next to Llama-, Qwen- or SEA-LION-scale systems on general reasoning, the base checkpoints need further fine-tuning before they're useful, and independent benchmarking outside Mesolitica's own leaderboard is scarce. Adoption numbers aren't public, so beyond the local NLP community the footprint is hard to quantify.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including MaLLaM's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with MaLLaM unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to MaLLaM directly →
Spotted something out of date? Suggest an update →
Alternatives to MaLLaM
More in LLMs & Chat
