Home › Best › cheapest LLM to run locally on your own GPU The cheapest open LLMs to run on your own hardware Open LLMs ranked by how little VRAM they need to run locally — the smaller the footprint, the cheaper the GPU you need and the closer to truly $0 it gets. Sorted by computed VRAM-to-run (lowest first), with the honest local and rent-a-GPU cost for each.
How this is ranked: Objective: ranks by engine-computed VRAM (proxy for self-hosting cost — lowest VRAM = cheapest hardware to buy/rent). All runs are $0-markup. Not a quality ranking; the user judges which cheap-to-run model is good enough.
1. Nomic Embed Text v1.5 — Nomic AI, 137M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK2. gemma 3 270m — google, 300M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK3. Qwen2.5 0.5B Instruct — Qwen, 500M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK4. Qwen2.5 0.5B — Qwen, 500M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK5. TinyLlama 1.1B Chat v1.0 — TinyLlama, 1.1B · ~2.0 GB VRAM · $0.11/1M API est. · commercial OK6. gpt2 large — openai-community, 800M · ~2.0 GB VRAM · $0.11/1M API est. · commercial OK7. h2ovl mississippi 800m — h2oai, 800M · ~2.0 GB VRAM · $0.11/1M API est. · commercial OK8. Qwen2 0.5B — Qwen, 500M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK9. Qwen2 0.5B Instruct — Qwen, 500M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK10. bloom 560m — bigscience, 600M · ~2.0 GB VRAM · $0.10/1M API est. · non-commercial11. gpt2 medium — openai-community, 400M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK12. SmolLM2 360M Instruct — HuggingFaceTB, 400M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK13. LLaMmlein 1B prerelease — LSX-UniWue, 1.1B · ~2.0 GB VRAM · $0.11/1M API est. · non-commercial14. t5gemma s s prefixlm — google, 300M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK15. bloomz 560m — bigscience, 600M · ~2.0 GB VRAM · $0.10/1M API est. · non-commercial16. MiniCPM5 1B — openbmb, 1.1B · ~2.0 GB VRAM · $0.11/1M API est. · commercial OK17. gemma 3 270m it — google, 300M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK18. pythia 410m — EleutherAI, 500M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK19. functiongemma 270m it — google, 300M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK20. vlt5 base keywords — Voicelab, 300M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK21. Falcon H1 0.5B Base — tiiuae, 500M · ~2.0 GB VRAM · $0.10/1M API est. · non-commercial22. Qwen2.5 Coder 0.5B Instruct — Qwen, 500M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK23. pythia 1b — EleutherAI, 1.1B · ~2.0 GB VRAM · $0.11/1M API est. · commercial OK24. MiniCPM4 0.5B — openbmb, 400M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK25. qwen sft countdown defaultproj — asingh15, 500M · ~2.0 GB VRAM · $0.10/1M API est. · non-commercial26. Qwen3.6 35B A3B DFlash — z-lab, 500M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK27. pythia 410m deduped — EleutherAI, 500M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK28. Qwen3 8B DFlash b16 — z-lab, 1B · ~2.0 GB VRAM · $0.11/1M API est. · commercial OK29. SmolLM2 360M — HuggingFaceTB, 400M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK30. LFM2.5 350M — LiquidAI, 400M · ~2.0 GB VRAM · $0.10/1M API est. · non-commercial31. tinyllama oneshot w8w8 test static shape change — nm-testing, 1.1B · ~2.0 GB VRAM · $0.11/1M API est. · non-commercial32. TinyLlama 1.1B intermediate step 1431k 3T — TinyLlama, 1.1B · ~2.0 GB VRAM · $0.11/1M API est. · commercial OK33. gpt oss 20b speculator.eagle3 — RedHatAI, 900M · ~2.0 GB VRAM · $0.11/1M API est. · commercial OK34. LFM2.5 350M Base — LiquidAI, 400M · ~2.0 GB VRAM · $0.10/1M API est. · non-commercial35. NVIDIA Nemotron 3.5 Lightning 30B A3B NVFP4 DSpark — nvidia, 800M · ~2.0 GB VRAM · $0.11/1M API est. · non-commercial36. gemma 4 26B A4B it DFlash — z-lab, 400M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK37. TIPO 500M — KBlueLeaf, 500M · ~2.0 GB VRAM · $0.10/1M API est. · non-commercial38. Qwen3.5 122B A10B DFlash — z-lab, 800M · ~2.0 GB VRAM · $0.11/1M API est. · commercial OK39. Qwen3.5 4B DFlash — z-lab, 600M · ~2.0 GB VRAM · $0.10/1M API est. · commercial OK40. MiniCPM5 1B SFT — openbmb, 1.1B · ~2.0 GB VRAM · $0.11/1M API est. · commercial OKShowing the top 40 of 604. See all →
More: all "best" lists · Outcome Lab · all models
Open the free LLM cost calculator → · $0 markup. © 2026 Cynosure LLC.