MobileLLM
Sub-billion-parameter LLMs designed for on-device use.
About
MobileLLM is Meta's ICML 2024 family of sub-billion-parameter language models (125M to 1.5B) built specifically for phones and edge devices. Instead of chasing scale, the team optimized the architecture — deep-and-thin layouts, embedding sharing, grouped-query attention — and the 125M/350M models beat prior same-scale state of the art by up to 4.3% on zero-shot commonsense reasoning.
These are research models for on-device experimentation, not assistants: no instruction tuning out of the box in the base releases, and the weights ship under Meta's custom Fair License rather than a standard open license. A reasoning follow-up, MobileLLM-R1, arrived in late 2025.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- 2.7%/4.3% accuracy gain over prior 125M/350M SoTA on zero-shot commonsense reasoning (paper-reported)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Meta Research, ICML 2024; training code public on GitHub; HF weights release Oct 2024; Influential architecture work for sub-billion LLMs; reasoning follow-up (MobileLLM-R1) appeared late 2025; Custom Fair License, not OSI-open; base models lack instruction tuning; Research models for on-device experimentation, not assistants. License: Meta Fair License (custom) — not OSI-open, check commercial terms.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 6.0/10
- Maintenance · recency, releases, issue load
- 6.0/10
- Openness · license + docs
- 7.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Meta Research, ICML 2024; training code public on GitHub; HF weights release Oct 2024
- · Influential architecture work for sub-billion LLMs; reasoning follow-up (MobileLLM-R1) appeared late 2025
- · Custom Fair License, not OSI-open; base models lack instruction tuning
- · Research models for on-device experimentation, not assistants