The best small open LLMs (under 4B parameters)

Every open LLM under 4 billion parameters in our catalog — the tiny, fast, edge-friendly tier that runs almost anywhere, including CPUs and phones. Ranked by popularity/recognition with the honest $0 cost to run each locally. You judge which small model is sharpest for your task.

How this is ranked: Objective size cut (under 4B is a catalog fact). Within the tier we sort by popularity (a real signal: HF downloads/recognition), explicitly framed as 'most-recognized small models — judge quality yourself,' not 'this tiny model is best.'

1. Qwen3 0.6B — Qwen, 800M · ~3.0 GB VRAM · $0.11/1M API est. · commercial OK
2. Qwen2.5 3B Instruct — Qwen, 3.1B · ~4.0 GB VRAM · $0.12/1M API est. · non-commercial
3. Llama 3.2 3B Instruct — Meta, 3B · ~5.0 GB VRAM · $0.12/1M API est. · commercial OK
4. Qwen2.5 1.5B Instruct — Qwen, 1.5B · ~3.0 GB VRAM · $0.11/1M API est. · commercial OK
5. gemma 3 270m — google, 300M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK
6. Qwen3 1.7B — Qwen, 2B · ~5.0 GB VRAM · $0.12/1M API est. · commercial OK
7. BGE-M3 — BAAI, 567M · ~3.0 GB VRAM · $0.10/1M API est. · commercial OK
8. Qwen2.5 0.5B Instruct — Qwen, 500M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK
9. Qwen2 1.5B Instruct — Qwen, 1.5B · ~3.0 GB VRAM · $0.11/1M API est. · commercial OK
10. Llama 3.2 1B Instruct — Meta, 1.2B · ~3.0 GB VRAM · $0.11/1M API est. · commercial OK
11. Llama 3.2 1B — meta-llama, 1.2B · ~3.0 GB VRAM · $0.11/1M API est. · commercial OK
12. Qwen2.5 0.5B — Qwen, 500M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK
13. Phi-3.5-mini Instruct — Microsoft, 3.8B · ~5.0 GB VRAM · $0.13/1M API est. · commercial OK
14. gemma 3 1b it — google, 1B · ~3.0 GB VRAM · $0.11/1M API est. · commercial OK
15. TinyLlama 1.1B Chat v1.0 — TinyLlama, 1.1B · ~2.0 GB VRAM · $0.11/1M API est. · commercial OK
16. gpt2 large — openai-community, 800M · ~2.0 GB VRAM · $0.11/1M API est. · commercial OK
17. OpenELM 1 1B Instruct — apple, 1.1B · ~3.0 GB VRAM · $0.11/1M API est. · non-commercial
18. PowerMoE 3b — ibm-research, 3.4B · ~4.0 GB VRAM · $0.13/1M API est. · commercial OK
19. Phi 4 mini instruct — microsoft, 3.8B · ~6.0 GB VRAM · $0.13/1M API est. · commercial OK
20. Qwen2.5 1.5B — Qwen, 1.5B · ~3.0 GB VRAM · $0.11/1M API est. · commercial OK
21. h2ovl mississippi 800m — h2oai, 800M · ~2.0 GB VRAM · $0.11/1M API est. · commercial OK
22. h2ovl mississippi 2b — h2oai, 2.2B · ~4.0 GB VRAM · $0.12/1M API est. · commercial OK
23. Nomic Embed Text v1.5 — Nomic AI, 137M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK
24. Qwen2 0.5B — Qwen, 500M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK
25. SmolLM 1.7B Instruct quantized.w4a16 — nm-testing, 1.8B · ~3.0 GB VRAM · $0.11/1M API est. · commercial OK
26. Qwen2.5 1.5B quantized.w8a8 — RedHatAI, 1.8B · ~3.0 GB VRAM · $0.11/1M API est. · commercial OK
27. Qwen2.5 Math 1.5B — Qwen, 1.5B · ~3.0 GB VRAM · $0.11/1M API est. · commercial OK
28. gpt neo 2.7B — EleutherAI, 2.7B · ~4.0 GB VRAM · $0.12/1M API est. · commercial OK
29. Llama 3.2 3B — meta-llama, 3.2B · ~5.0 GB VRAM · $0.13/1M API est. · commercial OK
30. Qwen2 0.5B Instruct — Qwen, 500M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK
31. Phi tiny MoE instruct — microsoft, 3.8B · ~4.0 GB VRAM · $0.13/1M API est. · commercial OK
32. Qwen2.5 Coder 1.5B Instruct — Qwen, 1.5B · ~3.0 GB VRAM · $0.11/1M API est. · commercial OK
33. Qwen2.5 Coder 3B — Qwen, 3.1B · ~4.0 GB VRAM · $0.12/1M API est. · non-commercial
34. Qwen3 1.7B Base — Qwen, 1.7B · ~4.0 GB VRAM · $0.11/1M API est. · commercial OK
35. DeepSeek R1 Distill Qwen 1.5B — deepseek-ai, 1.8B · ~3.0 GB VRAM · $0.11/1M API est. · commercial OK
36. Phi 3 mini 4k instruct — microsoft, 3.8B · ~5.0 GB VRAM · $0.13/1M API est. · commercial OK
37. Qwen3 0.6B Base — Qwen, 600M · ~3.0 GB VRAM · $0.10/1M API est. · commercial OK
38. SmolLM3 3B — HuggingFaceTB, 3.1B · ~5.0 GB VRAM · $0.12/1M API est. · commercial OK
39. Qwen2.5 3B — Qwen, 3.1B · ~4.0 GB VRAM · $0.12/1M API est. · non-commercial
40. bloom 560m — bigscience, 600M · ~2.0 GB VRAM · $0.10/1M API est. · non-commercial

Showing the top 40 of 113. See all →

More: all "best" lists · cost calculator · all models