RightNow-Arabic-0.5B-Turbo: An Open Sub-1B Arabic Language Model via Vocabulary Injection and Edge-First Deployment

arXiv:2605.28827v1 Announce Type: cross Abstract: Open Arabic large language models split into two classes: sub-1B multilingual models that treat Arabic as an afterthought (Qwen2.5-0.5B, Falcon-H1-0.5B), and 7B-70B Arabic-specialized models that require a server to run (Jais, AceGPT, ALLaM, SILMA). The one published attempt at a sub-2B Arabic-specialized model, Kuwain-1.5B, never released its weights. We present RightNow-Arabic-0.5B-Turbo, a 518M-parameter Arabic-specialized decoder LLM built on Qwen2.5-0.5B. The pipeline adds 27,032 Arabic tokens via mean-subtoken initialization, continues pr
The continuous evolution of open-source models, combined with increasing demand for localized AI solutions, drives the development of specialized sub-1B Arabic LLMs for edge deployment.
This development addresses the critical need for performant, accessible, and culturally relevant AI tools in non-dominant languages, potentially enabling broader AI adoption and innovation outside major tech hubs.
The availability of open-source, specialized, sub-1B Arabic LLMs for edge deployment changes the landscape for small-to-medium enterprises and developers seeking high-performance local AI solutions without relying on large server infrastructure.
- · Arabic-speaking developers
- · Edge AI hardware manufacturers
- · Startups in MENA region
- · Open-source AI community
- · Large-scale proprietary Arabic LLM providers
- · Cloud-reliant AI service providers
- · Developers solely focused on large, generic multilingual models
Increased availability and performance of localized Arabic AI applications on consumer devices and embedded systems.
Accelerated innovation and competitive differentiation within the Arabic-speaking AI ecosystem, fostering new businesses and use cases.
Reduced dependence on foreign AI infrastructure and models, contributing to digital sovereignty and cultural preservation through AI.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG