AI models that run on 8 GB of VRAM

173 open models fit in 8 GB of VRAM at their default quant — enough for an entry GPU like an RTX 3060 / 4060 or 8 GB laptop card. Most capable first; run any of them for $0 on your own hardware.

• internlm3 8b instruct — 8.8B, internlm · ~8.0 GB VRAM
• Nemotron Labs Diffusion 8B Base — 8.5B, nvidia · ~7.0 GB VRAM
• LFM2.5 8B A1B — 8.5B, LiquidAI · ~7.0 GB VRAM
• granite 3.0 8b instruct — 8.2B, ibm-granite · ~7.0 GB VRAM
• Llama 3.1 8B Instruct — 8B, Meta · ~8.0 GB VRAM
• Qwen2-VL 7B Instruct — 8B, Alibaba · ~7.0 GB VRAM
• Llama 3.1 8B Instruct (Abliterated) — 8B, mlabonne (community) · ~8.0 GB VRAM
• Hermes 3 — Llama 3.1 8B — 8B, Nous Research · ~8.0 GB VRAM
• Dolphin 3.0 — Llama 3.1 8B — 8B, Cognitive Computations · ~8.0 GB VRAM
• LLaDA 8B Instruct — 8B, GSAI-ML · ~8.0 GB VRAM
• DeepSeek R1 Distill Llama 8B — 8B, deepseek-ai · ~8.0 GB VRAM
• saiga llama3 8b — 8B, IlyaGusev · ~7.0 GB VRAM
• Meta Llama 3.1 8B Instruct — 8B, unsloth · ~8.0 GB VRAM
• LLaDA 1.5 — 8B, GSAI-ML · ~8.0 GB VRAM
• Meta Llama 3 8B Instruct — 8B, NousResearch · ~7.0 GB VRAM
• Meta Llama 3.1 8B Instruct — 8B, NousResearch · ~8.0 GB VRAM
• Llama 3.1 8B Instruct — 8B, unsloth · ~8.0 GB VRAM
• LLaDA 8B Base — 8B, GSAI-ML · ~8.0 GB VRAM
• Humanish Roleplay Llama 3.1 8B — 8B, vicgalle · ~8.0 GB VRAM
• llava onevision qwen2 7b ov — 8B, lmms-lab · ~7.0 GB VRAM
• Llama 3.1 Nemotron Safety Guard 8B v3 — 8B, nvidia · ~8.0 GB VRAM
• NeuralDaredevil 8B abliterated — 8B, mlabonne · ~7.0 GB VRAM
• L3 8B Stheno v3.2 — 8B, Sao10K · ~7.0 GB VRAM
• EXAONE 3.5 7.8B Instruct — 7.8B, LGAI-EXAONE · ~8.0 GB VRAM
• hf moshiko — 7.8B, kmhf · ~8.0 GB VRAM
• internlm2 5 7b chat — 7.7B, internlm · ~8.0 GB VRAM
• internlm2 chat 7b — 7.7B, internlm · ~8.0 GB VRAM
• Qwen2 7B Instruct — 7.6B, Qwen · ~7.0 GB VRAM
• Qwen2.5 7B — 7.6B, Qwen · ~7.0 GB VRAM
• DeepSeek R1 Distill Qwen 7B — 7.6B, deepseek-ai · ~7.0 GB VRAM
• Dream v0 Instruct 7B — 7.6B, Dream-org · ~7.0 GB VRAM
• Qwen2.5 Coder 7B — 7.6B, Qwen · ~7.0 GB VRAM
• Phi mini MoE instruct — 7.6B, microsoft · ~7.0 GB VRAM
• Qwen2.5 Math 7B Instruct — 7.6B, Qwen · ~6.0 GB VRAM
• Qwen2.5 7B Instruct — 7.6B, unsloth · ~7.0 GB VRAM
• OLMo 2 1124 7B Instruct — 7.3B, allenai · ~8.0 GB VRAM
• Mistral 7B Instruct v0.3 — 7.2B, Mistral AI · ~8.0 GB VRAM
• Mistral 7B Instruct v0.2 — 7.2B, mistralai · ~8.0 GB VRAM
• Mistral 7B v0.1 — 7.2B, mistralai · ~8.0 GB VRAM
• Mistral 7B Instruct v0.1 — 7.2B, mistralai · ~8.0 GB VRAM
• zephyr 7b beta — 7.2B, HuggingFaceH4 · ~8.0 GB VRAM
• Qwen2.5 7B Instruct — 7B, Alibaba · ~7.0 GB VRAM
• Qwen2.5-Coder 7B Instruct — 7B, Alibaba · ~7.0 GB VRAM
• Qwen2.5 7B Instruct (Abliterated) — 7B, huihui-ai (community) · ~7.0 GB VRAM
• deepseek coder 7b instruct v1.5 — 6.9B, deepseek-ai · ~8.0 GB VRAM
• granite 4.0 h tiny — 6.9B, ibm-granite · ~7.0 GB VRAM
• OLMoE 1B 7B 0125 Instruct — 6.9B, allenai · ~6.0 GB VRAM
• OLMoE 1B 7B 0924 — 6.9B, allenai · ~6.0 GB VRAM
• llama 7b — 6.7B, huggyllama · ~7.0 GB VRAM
• granite 4.0 tiny preview — 6.7B, ibm-granite · ~7.0 GB VRAM
• Llama 2 7b hf — 6.7B, NousResearch · ~8.0 GB VRAM
• Llama 2 7b chat hf — 6.7B, NousResearch · ~8.0 GB VRAM
• Gemma 4 E4B it NVFP4 — 6B, bg-digitalservices · ~8.0 GB VRAM
• Qwen3.6 27B MTPLX Optimized Speed — 4.7B, Youssofal · ~7.0 GB VRAM
• Josiefied Qwen3 VL 4B Instruct abliterated beta v1 — 4.4B, Goekdeniz-Guelmez · ~6.0 GB VRAM
• Qwen3 4B — 4B, Qwen · ~5.0 GB VRAM
• Qwen3 4B Instruct 2507 — 4B, Qwen · ~5.0 GB VRAM
• Rio 3.0 Open Mini — 4B, prefeitura-rio · ~5.0 GB VRAM
• NVIDIA Nemotron 3 Nano 4B BF16 — 4B, nvidia · ~6.0 GB VRAM
• Qwen3 4B Base — 4B, Qwen · ~5.0 GB VRAM
• Qwen3 4B Thinking 2507 — 4B, Qwen · ~5.0 GB VRAM
• Nanbeige4.1 3B — 3.9B, Nanbeige · ~5.0 GB VRAM
• Phi-3.5-mini Instruct — 3.8B, Microsoft · ~5.0 GB VRAM
• Phi 4 mini instruct — 3.8B, microsoft · ~6.0 GB VRAM
• Phi tiny MoE instruct — 3.8B, microsoft · ~4.0 GB VRAM
• Phi 3 mini 4k instruct — 3.8B, microsoft · ~5.0 GB VRAM
• Nemotron Labs Diffusion 3B Base — 3.8B, nvidia · ~4.0 GB VRAM
• Phi 4 mini reasoning — 3.8B, microsoft · ~6.0 GB VRAM
• Nemotron Labs Diffusion 3B — 3.8B, nvidia · ~5.0 GB VRAM
• HyperCLOVAX SEED Vision Instruct 3B — 3.7B, naver-hyperclovax · ~5.0 GB VRAM
• PowerLM 3b — 3.5B, ibm-research · ~5.0 GB VRAM
• PowerMoE 3b — 3.4B, ibm-research · ~4.0 GB VRAM
• granite 4.1 3b — 3.4B, ibm-granite · ~5.0 GB VRAM
• granite 4.0 micro — 3.4B, ibm-granite · ~5.0 GB VRAM
• tiny aya base — 3.3B, CohereLabs · ~5.0 GB VRAM
• tiny aya global — 3.3B, CohereLabs · ~5.0 GB VRAM
• Llama 3.2 3B — 3.2B, meta-llama · ~5.0 GB VRAM
• Llama 3.2 3B Instruct — 3.2B, unsloth · ~5.0 GB VRAM
• Llama 3.2 3B Instruct pythonic — 3.2B, baseten · ~5.0 GB VRAM
• Qwen2.5 3B Instruct — 3.1B, Qwen · ~4.0 GB VRAM
• Qwen2.5 Coder 3B — 3.1B, Qwen · ~4.0 GB VRAM
• SmolLM3 3B — 3.1B, HuggingFaceTB · ~5.0 GB VRAM
• Qwen2.5 3B — 3.1B, Qwen · ~4.0 GB VRAM
• SmolLM3 3B Base — 3.1B, HuggingFaceTB · ~5.0 GB VRAM
• Qwen2.5 Coder 3B Instruct — 3.1B, Qwen · ~4.0 GB VRAM
• Llama 3.2 3B Instruct — 3B, Meta · ~5.0 GB VRAM
• starcoder2 3b — 3B, bigcode · ~4.0 GB VRAM
• kimi k2.6 eagle3 mla — 3B, lightseekorg · ~4.0 GB VRAM
• phi 2 — 2.8B, microsoft · ~4.0 GB VRAM
• stablelm 3b 4e1t — 2.8B, stabilityai · ~5.0 GB VRAM

Other GPU budgets

16 GB of VRAM · 24 GB of VRAM · 48 GB of VRAM · All models

Open the free advisor → · Prices as of 2026-06-17. We're an honest advisor — $0 markup, your own accounts, we never resell compute. © 2026 Cynosure LLC.