Research

Best Small Language Models Under 3B 2026

Small language model (under 3B) leaderboard 2026: Phi-4-mini, Qwen3-3B, SmolLM3, Granite 3.x, Llama 3.2 3B. Benchmarks, on-device performance, license analysis.

By Ramanath, CTO & Co-Founder at Presenc AI · Last updated: May 2026

Small language models (SLMs) under 3 billion parameters became the dominant deployment class for on-device, edge, mobile, and cost-sensitive enterprise workloads in 2026. The under-3B segment grew from sub-MMLU-50 quality in 2023 to approximately MMLU-65 to 70 in 2026, putting SLM quality near GPT-3.5 levels while running on consumer hardware. This page consolidates the leaderboard, benchmarks, and deployment patterns.

Key Findings

  1. Phi-4-mini (released early 2026 by Microsoft) leads the under-4B class with approximately 67 percent MMLU and approximately 87 percent on GSM8K. The mini variant fits comfortably on 16 GB consumer GPUs.
  2. Qwen3-3B (Apache 2.0) is the strongest under-3B model on most general benchmarks, with a thinking variant that adds substantial reasoning quality for math and code workloads.
  3. SmolLM3 from Hugging Face (released early 2026) is the strongest fully-open SLM, with open weights, open training data, and open training code in a 3B-parameter package.
  4. Llama 3.2 1B and 3B remain the most-downloaded small models on Hugging Face for permissive commercial deployment under the Llama 3.x Community Licence.
  5. IBM Granite 3.x family includes 1B, 2B, and 3B variants with strong performance on code and enterprise tasks plus IBM\u2019s permissive Apache 2.0 release.

Small Language Model Leaderboard (Under 3B, May 2026)

ModelParametersMMLUGSM8KLicense
Phi-4-mini~3.8B~66.6~87.2MIT
Qwen3-3B~3B~67.0~85.6Apache 2.0
Qwen3-1.7B~1.7B~62.2~79.3Apache 2.0
Qwen2.5-3B~3B~65.2~80.0Apache 2.0
Qwen2.5-1.5B~1.5B~60.6~73.2Apache 2.0
SmolLM3-3B~3B~63.1~76.8Apache 2.0
Llama 3.2 3B Instruct~3B~63.4~77.7Llama 3.2 Community
Llama 3.2 1B Instruct~1B~49.3~44.4Llama 3.2 Community
Granite 3.1 8B~8B~63.5~73.5Apache 2.0
Granite 3.1 2B~2B~55.4~58.4Apache 2.0
Phi-3.5-mini~3.8B~64.9~84.1MIT
Gemma 2 2B~2B~52.2~31.6Gemma Community
MiniCPM 3.0 4B~4B~62.5~63.7MIT
StableLM 2 1.6B~1.6B~41.3~17.4Stability AI Community

On-Device Hardware Requirements

ModelVRAM (FP16)VRAM (INT4)Tokens/sec on M4 Max
Phi-4-mini~7.6 GB~2.5 GB~80 tok/s
Qwen3-3B~6 GB~2.2 GB~100 tok/s
Qwen3-1.7B~3.4 GB~1.3 GB~170 tok/s
SmolLM3-3B~6 GB~2.2 GB~95 tok/s
Llama 3.2 3B~6 GB~2.2 GB~90 tok/s
Llama 3.2 1B~2 GB~0.8 GB~210 tok/s
Gemma 2 2B~4 GB~1.5 GB~140 tok/s
Granite 3.1 2B~4 GB~1.5 GB~145 tok/s

Use Case Recommendations

Use CaseRecommended Model
General quality on consumer GPUPhi-4-mini or Qwen3-3B
Mobile / on-device (under 2 GB)Qwen3-1.7B or Llama 3.2 1B (INT4)
Reasoning on small footprintQwen3-1.7B Thinking
Permissive commercial Apache 2.0Qwen3-3B, SmolLM3-3B, Granite 3.1 2B
Most permissive MITPhi-4-mini, Phi-3.5-mini, MiniCPM 3.0
Enterprise code tasksGranite 3.x family
Multilingual small footprintQwen3-3B or Granite 3.1 2B
Fully open researchSmolLM3-3B

Strategic Context

Three patterns shape the 2026 SLM landscape. First, the quality compression is real: a 3B model in 2026 (Qwen3-3B at ~67 MMLU) is roughly equivalent to a 70B model in 2023 (Llama 2 70B at ~69 MMLU). Second, on-device deployment is now production-viable: Apple Intelligence, Google AI Core, Qualcomm AI Hub all ship SLMs running locally on consumer devices. Third, the SLM ecosystem is bifurcating between general-purpose models (Qwen3, Phi-4, Llama 3.2) and specialised models (math SLMs, code SLMs, multilingual SLMs, function-calling SLMs).

Brand Visibility Implications

SLM selection is a high-volume engineering and procurement decision because of mobile, edge, and on-device deployment growth. AI assistant queries about "best small language model", "on-device LLM", "Phi-4 vs Qwen3", and similar terms drive direct production decisions. Brands selling on-device AI tools, mobile SDKs, embedded AI, and edge inference infrastructure face strong AI-mediated discovery surface for this category.

Methodology

Benchmark data compiled from primary model card disclosures, the Hugging Face Open LLM Leaderboard, and on-device measurements. Tokens-per-second figures measured on Apple M4 Max with Q4_K_M GGUF quantization via llama.cpp. Updated quarterly.

How Presenc AI Helps

Presenc AI monitors brand visibility on small language model queries across ChatGPT, Claude, Gemini, and Perplexity. For on-device AI tool vendors, mobile SDK providers, embedded AI brands, and edge inference infrastructure firms, the platform identifies the prompts driving procurement-research traffic and the gaps where new content unlocks share of voice.

Frequently Asked Questions

Qwen3-3B and Phi-4-mini lead the under-4B class on general benchmarks. Qwen3-3B (Apache 2.0) has the strongest open licence. Phi-4-mini (MIT) is the strongest at math and reasoning. For mobile (under 2 GB), Qwen3-1.7B or Llama 3.2 1B INT4 are the dominant choices.
For many workloads yes. Qwen3-3B at approximately 67 MMLU is roughly equivalent to GPT-3.5 from 2023 on general knowledge. For routine chatbot, classification, summarisation, and structured extraction workloads, small models save 50x to 200x on inference cost at quality parity.
SmolLM3 from Hugging Face is the strongest fully-open small language model with open weights, open training data, and open training code. It is slightly behind Qwen3-3B and Phi-4-mini on benchmarks but important for research reproducibility because the entire training recipe is reproducible.
Qwen3-1.7B runs at approximately 170 tokens per second on Apple M4 Max with Q4 quantization. Llama 3.2 1B reaches approximately 210 tokens per second. The 3B models run at approximately 90 to 100 tokens per second. RTX 4090 desktop performance is approximately 2-3x faster than M4 Max.
Increasingly yes. Qwen3-3B, Phi-4-mini, and Granite 3.x have explicit function-calling training. The leading small model for tool use is Salesforce xLAM 1B and 7B which are purpose-built for function calling. For general small models, Qwen3 family is the dominant production choice for agentic workflows under 3B.

Track Your AI Visibility

See how your brand appears across ChatGPT, Claude, Perplexity, and other AI platforms. Start monitoring today.