Quality × Hardware
Open-source AI leaderboard
Models ranked by composite quality from public benchmarks. Pair this with your hardware grade — pick the smartest model that actually fits.
Quality × VRAM matrix
Each cell lists the strongest models in that quality and VRAM band. Top-left = smartest. Bottom-right = biggest fit.
| Quality \ VRAM | ≤4 GB | 4–8 GB | 8–12 GB | 12–16 GB | 16–24 GB | 24–48 GB | 48 GB+ |
|---|---|---|---|---|---|---|---|
| Excellent (80+) | |||||||
| Strong (65–80) | — | ||||||
| Capable (50–65) | — | — | — | — | |||
| Light (35–50) | — | — | — | — | |||
| Tiny (<35) | — | — | — | — | — |
Full ranking (79 models)
| # | Model | Composite | MMLU | HumanEval | GSM8K | Arena ELO | Min VRAM |
|---|---|---|---|---|---|---|---|
| 1 | Qwen3 235B-A22B Alibaba · 235B · llm | 92.7 | 87.4 | 92.7 | 96.8 | 1310 | 144 GB |
| 2 | Qwen 2.5 Coder 14B Alibaba · 14B · code | 89.7 | — | 89.7 | — | — | 8.87 GB |
| 3 | Qwen 2.5 32B Alibaba · 32B · llm | 87.8 | 83.3 | 88.4 | 95.9 | 1235 | 18.99 GB |
| 4 | Llama 3.1 70B Instruct Meta · 70B · llm | 85.5 | 79.5 | 80.5 | 95.1 | 1248 | 40.1 GB |
| 5 | Yi Coder 9B 01.AI · 9B · code | 85.4 | — | 85.4 | — | — | 5.46 GB |
| 6 | Qwen3 30B-A3B Alibaba · 30.5B · llm | 85.3 | 81.5 | 86.4 | 95.0 | 1240 | 20 GB |
| 7 | Qwen 2.5 Coder 3B Alibaba · 3B · code | 84.1 | — | 84.1 | — | — | 2.46 GB |
| 8 | Qwen 2.5 14B Alibaba · 14B · llm | 83.6 | 79.7 | 83.5 | 94.2 | 1208 | 8.87 GB |
| 9 | Gemma 3 27B Google · 27B · llm | 83.3 | 76.9 | 87.8 | 89.0 | 1218 | 15.91 GB |
| 10 | DeepSeek R1 Distill 8B DeepSeek · 8B · llm | 80.0 | 70.5 | 80.5 | — | — | 5.08 GB |
| 11 | Qwen 2.5 7B Instruct Alibaba · 7.6B · llm | 79.8 | 74.2 | 84.8 | 91.6 | 1175 | 5.3 GB |
| 12 | Phi-4 Microsoft · 14B · llm | 79.7 | 84.8 | 82.6 | 95.4 | — | 8.93 GB |
| 13 | Phi-3.5 MoE Microsoft · 41.9B · llm | 79.4 | 78.9 | 70.7 | 88.7 | — | 24.11 GB |
| 14 | InternLM 2.5 7B Shanghai AI Lab · 7.7B · llm | 79.4 | 72.8 | — | 86.0 | — | 4.89 GB |
| 15 | DeepSeek Coder 6.7B DeepSeek · 6.7B · code | 78.6 | — | 78.6 | — | — | 4.3 GB |
| 16 | Mixtral 8x22B Instruct Mistral AI · 141B · llm | 78.6 | 77.8 | 76.5 | 88.4 | 1186 | 88 GB |
| 17 | Qwen 2.5 Coder 7B Alibaba · 7.6B · code | 78.0 | 67.6 | 88.4 | — | — | 4.86 GB |
| 18 | Phi-4 Mini 3.8B Microsoft · 3.8B · llm | 76.8 | 67.3 | 74.4 | 88.6 | — | 2.82 GB |
| 19 | DeepSeek R1 Distill 1.5B DeepSeek · 1.5B · llm | 74.9 | — | 65.9 | — | — | 1.54 GB |
| 20 | Falcon 3 10B TII · 10B · llm | 73.1 | 73.1 | — | — | — | 6.36 GB |
| 21 | Phi-3.5 Mini 3.8B Microsoft · 3.8B · llm | 72.6 | 68.9 | 62.8 | 86.2 | — | 2.73 GB |
| 22 | EXAONE 3.5 7.8B LG AI · 7.8B · llm | 71.1 | 65.4 | 76.8 | — | — | 4.94 GB |
| 23 | Qwen 2.5 Coder 1.5B Alibaba · 1.5B · code | 70.7 | — | 70.7 | — | — | 1.54 GB |
| 24 | MiniCPM-V 2.6 OpenBMB · 2B · multimodal | 69.7 | — | — | — | — | 2.1 GB |
| 25 | Yi 1.5 9B Chat 01.AI · 9B · llm | 69.7 | 69.7 | — | — | — | 5.46 GB |
| 26 | Qwen 2.5 3B Alibaba · 3B · llm | 68.9 | 65.6 | 74.4 | 86.7 | 1095 | 2.46 GB |
| 27 | Gemma 3 12B Google · 12B · llm | 68.5 | 68.7 | 68.3 | — | — | 7.3 GB |
| 28 | Mistral Small 22B Mistral AI · 22B · llm | 68.4 | 73.0 | 70.2 | — | 1148 | 12.93 GB |
| 29 | Granite 3.3 8B IBM · 8B · llm | 68.1 | 64.7 | 71.4 | — | — | 5.1 GB |
| 30 | Falcon 3 7B TII · 7B · llm | 67.4 | 67.4 | — | — | — | 5 GB |
| 31 | Llama 3.1 8B Instruct Meta · 8B · llm | 66.2 | 68.4 | 62.2 | 80.5 | 1115 | 5.08 GB |
| 32 | Qwen 2.5 1.5B Alibaba · 1.5B · llm | 65.2 | 60.9 | 61.6 | 73.2 | — | 1.54 GB |
| 33 | DeepSeek Coder 1.3B DeepSeek · 1.3B · code | 65.2 | — | 65.2 | — | — | 1.31 GB |
| 34 | Gemma 2 9B Instruct Google · 9.2B · llm | 65.2 | 71.3 | 40.2 | 76.7 | 1190 | 5.87 GB |
| 35 | Yi 1.5 6B Chat 01.AI · 6B · llm | 64.1 | 64.1 | — | — | — | 3.92 GB |
| 36 | Gemma 3 MoE 9B Google · 9B · llm | 63.0 | 70.2 | 55.9 | — | — | 7 GB |
| 37 | Mistral Nemo 12B Mistral AI · 12B · llm | 62.4 | 68.0 | 56.7 | — | — | 7.46 GB |
| 38 | Qwen 2.5 Coder 0.5B Alibaba · 0.5B · code | 61.6 | — | 61.6 | — | — | 1.13 GB |
| 39 | Mixtral 8x7B Instruct Mistral AI · 46.7B · llm | 60.8 | 70.6 | 40.2 | 74.4 | 1132 | 25.13 GB |
| 40 | OpenChat 3.5 7B OpenChat · 7B · llm | 60.6 | 65.8 | 55.5 | — | — | 4.57 GB |
| 41 | OLMo 2 7B Allen AI · 7B · llm | 60.4 | 60.4 | — | — | — | 4.67 GB |
| 42 | Phi-3.5 Vision Microsoft · 4.2B · multimodal | 60.2 | — | — | — | — | 3.2 GB |
| 43 | Solar 10.7B Upstage · 10.7B · llm | 60.0 | 65.9 | — | — | 1116 | 6.52 GB |
| 44 | Qwen2-VL 2B Alibaba · 2.2B · multimodal | 57.5 | — | — | — | — | 1.42 GB |
| 45 | EXAONE 3.5 2.4B LG AI · 2.4B · llm | 56.4 | 59.7 | 53.1 | — | — | 2.03 GB |
| 46 | CodeGemma 7B Google · 8.5B · code | 56.1 | — | 56.1 | — | — | 5.46 GB |
| 47 | Nemotron Mini 4B NVIDIA · 4B · llm | 56.1 | 56.1 | — | — | — | 3.01 GB |
| 48 | Falcon 3 3B TII · 3B · llm | 55.7 | 55.7 | — | — | — | 2.37 GB |
| 49 | Gemma 3 4B Google · 4B · llm | 54.7 | 58.1 | 51.2 | — | — | 2.82 GB |
| 50 | LLaVA 1.6 7B LLaVA · 7B · multimodal | 54.6 | 60.0 | — | — | — | 5 GB |
| 51 | Llama 3.2 3B Instruct Meta · 3.2B · llm | 54.2 | 63.4 | 35.0 | 77.7 | 1063 | 2.38 GB |
| 52 | Danube 3 4B H2O.ai · 4B · llm | 53.9 | 53.9 | — | — | — | 2.73 GB |
| 53 | Granite 3.3 2B IBM · 2B · llm | 52.4 | 52.4 | — | — | — | 1.94 GB |
| 54 | Granite 3.0 3B-A800M IBM · 3.4B · llm | 48.3 | 48.3 | — | — | — | 2.42 GB |
| 55 | Mistral 7B Instruct v0.3 Mistral AI · 7.3B · llm | 47.0 | 60.1 | 30.5 | 50.5 | — | 4.57 GB |
| 56 | StableLM Zephyr 3B Stability AI · 3B · llm | 46.0 | 46.0 | — | — | — | 2.09 GB |
| 57 | PaliGemma 3B Google · 3B · multimodal | 45.5 | — | — | — | — | 2.5 GB |
| 58 | Falcon 3 1B TII · 1B · llm | 42.9 | 42.9 | — | — | — | 1.48 GB |
| 59 | Moondream 2 Moondream · 1.8B · multimodal | 42.8 | — | — | — | — | 1.5 GB |
| 60 | SmolLM2 1.7B HuggingFace · 1.7B · llm | 41.8 | 51.4 | 32.3 | — | — | 1.48 GB |
| 61 | Yi Coder 1.5B 01.AI · 1.5B · code | 41.5 | — | 41.5 | — | — | 1.4 GB |
| 62 | Rocket 3B Pansophic · 3B · llm | 41.0 | 41.0 | — | — | — | 2.09 GB |
| 63 | Qwen 2.5 0.5B Alibaba · 0.5B · llm | 39.9 | 47.5 | 30.5 | 41.6 | — | 0.96 GB |
| 64 | OLMoE 1B-7B AI2 · 6.9B · llm | 38.7 | 54.1 | 23.8 | 38.2 | — | 4.42 GB |
| 65 | DeepSeek MoE 16B DeepSeek · 16.4B · llm | 37.9 | 49.0 | 26.8 | — | — | 11 GB |
| 66 | Gemma 2 2B Google · 2.6B · llm | 37.6 | 51.3 | 17.7 | 23.9 | 1130 | 2.09 GB |
| 67 | Code Llama 13B Instruct Meta · 13B · code | 36.0 | — | 36.0 | — | — | 7.83 GB |
| 68 | StarCoder2 7B BigCode · 7B · code | 35.4 | — | 35.4 | — | — | 4.66 GB |
| 69 | Llama 3.2 1B Instruct Meta · 1.24B · llm | 33.5 | 49.3 | 18.0 | 44.4 | 989 | 1.25 GB |
| 70 | Stable Code 3B Stability AI · 3B · code | 32.4 | — | 32.4 | — | — | 2.09 GB |
| 71 | Gemma 3 1B Google · 1B · llm | 31.7 | 38.8 | 24.6 | — | — | 1.25 GB |
| 72 | StarCoder2 3B BigCode · 3B · code | 31.7 | — | 31.7 | — | — | 2.26 GB |
| 73 | Code Llama 7B Meta · 7B · code | 31.7 | — | 31.7 | — | — | 4.3 GB |
| 74 | CodeGemma 2B Google · 2B · code | 31.1 | — | 31.1 | — | — | 2.02 GB |
| 75 | SmolLM2 135M HuggingFace · 0.135B · llm | 30.1 | 30.1 | — | — | — | 0.64 GB |
| 76 | Granite 3.0 1B-A400M IBM · 1.3B · llm | 28.5 | 28.5 | — | — | — | 1.27 GB |
| 77 | Danube 3 500M H2O.ai · 0.5B · llm | 28.4 | 28.4 | — | — | — | 0.8 GB |
| 78 | SmolLM2 360M HuggingFace · 0.36B · llm | 24.0 | 35.8 | 12.2 | — | — | 0.75 GB |
| 79 | TinyLlama 1.1B TinyLlama · 1.1B · llm | 15.5 | 25.5 | 5.5 | — | — | 1.12 GB |
Source: Public benchmarks aggregated from HF Open LLM Leaderboard, Chatbot Arena, BigCodeBench, MMLU-Pro, Math-500, and the original model technical reports. Numbers are best-effort April 2026 snapshots; some are reported by the model authors themselves and have not been independently verified. Last updated 2026-04-29.