Meta AI Released MobileLLM-R1: A Edge Reasoning Model with less than 1B Parameters and Achieves 2x–5x Performance Boost Over Other Fully Open-Source AI Models
Table of contents
Meta has launched MobileLLM-R1, a household of light-weight edge reasoning fashions now obtainable on Hugging Face. The launch contains fashions starting from 140M to 950M parameters, with a give attention to environment friendly mathematical, coding, and scientific reasoning at sub-billion scale.
Unlike general-purpose chat fashions, MobileLLM-R1 is designed for edge deployment, aiming to ship state-of-the-art reasoning accuracy whereas remaining computationally environment friendly.
What structure powers MobileLLM-R1?
The largest mannequin, MobileLLM-R1-950M, integrates a number of architectural optimizations:
- 22 Transformer layers with 24 consideration heads and 6 grouped KV heads.
- Embedding dimension: 1536; hidden dimension: 6144.
- Grouped-Query Attention (GQA) reduces compute and reminiscence.
- Block-wise weight sharing cuts parameter depend with out heavy latency penalties.
- SwiGLU activations enhance small-model illustration.
- Context size: 4K for base, 32K for post-trained fashions.
- 128K vocabulary with shared enter/output embeddings.
The emphasis is on lowering compute and reminiscence necessities, making it appropriate for deployment on constrained units.
How environment friendly is the coaching?
MobileLLM-R1 is notable for information effectivity:
- Trained on ~4.2T tokens in whole.
- By comparability, Qwen3’s 0.6B mannequin was skilled on 36T tokens.
- This means MobileLLM-R1 makes use of solely ≈11.7% of the info to achieve or surpass Qwen3’s accuracy.
- Post-training applies supervised fine-tuning on math, coding, and reasoning datasets.
This effectivity interprets straight into decrease coaching prices and useful resource calls for.
How does it carry out towards different open fashions?
On benchmarks, MobileLLM-R1-950M reveals vital positive aspects:
- MATH (MATH500 dataset): ~5× larger accuracy than Olmo-1.24B and ~2× larger accuracy than SmolLM2-1.7B.
- Reasoning and coding (GSM8K, AIME, LiveCodeBench): Matches or surpasses Qwen3-0.6B, regardless of utilizing far fewer tokens.
The mannequin delivers outcomes sometimes related with bigger architectures whereas sustaining a smaller footprint.
Where does MobileLLM-R1 fall quick?
The mannequin’s focus creates limitations:
- Strong in math, code, and structured reasoning.
- Weaker in normal dialog, commonsense, and inventive duties in comparison with bigger LLMs.
- Distributed beneath FAIR NC (non-commercial) license, which restricts utilization in manufacturing settings.
- Longer contexts (32K) increase KV-cache and reminiscence calls for at inference.
How does MobileLLM-R1 examine to Qwen3, SmolLM2, and OLMo?
Performance snapshot (post-trained fashions):
| Model | Params | Train tokens (T) | MATH500 | GSM8K | AIME’24 | AIME’25 | LiveCodeBench |
|---|---|---|---|---|---|---|---|
| MobileLLM-R1-950M | 0.949B | 4.2 | 74.0 | 67.5 | 15.5 | 16.3 | 19.9 |
| Qwen3-0.6B | 0.596B | 36.0 | 73.0 | 79.2 | 11.3 | 17.0 | 14.9 |
| SmolLM2-1.7B-Instruct | 1.71B | ~11.0 | 19.2 | 41.8 | 0.3 | 0.1 | 4.4 |
| OLMo-2-1B-Instruct | 1.48B | ~3.95 | 19.2 | 69.7 | 0.6 | 0.1 | 0.0 |
Key observations:
- R1-950M matches Qwen3-0.6B in math (74.0 vs 73.0) whereas requiring ~8.6× fewer tokens.
- Performance gaps vs SmolLM2 and OLMo are substantial throughout reasoning duties.
- Qwen3 maintains an edge in GSM8K, however the distinction is small in comparison with the coaching effectivity benefit.
Summary
Meta’s MobileLLM-R1 underscores a development towards smaller, domain-optimized fashions that ship aggressive reasoning with out huge coaching budgets. By reaching 2×–5× efficiency positive aspects over bigger open fashions whereas coaching on a fraction of the info, it demonstrates that effectivity—not simply scale—will outline the subsequent section of LLM deployment, particularly for math, coding, and scientific use circumstances on edge units.
Check out the Model on Hugging Face. Feel free to take a look at our GitHub Page for Tutorials, Codes and Notebooks. Also, be at liberty to observe us on Twitter and don’t neglect to affix our 100k+ ML SubReddit and Subscribe to our Newsletter.
The publish Meta AI Released MobileLLM-R1: A Edge Reasoning Model with less than 1B Parameters and Achieves 2x–5x Performance Boost Over Other Fully Open-Source AI Models appeared first on MarkTechPost.
