Loading...
Open Source LLM Rankings
Discover which open source large language models deliver the best quality per parameter. Models are ranked by efficiency score — benchmark quality divided by parameter count — using data from Artificial Analysis evaluations and HuggingFace metadata.
Explore hardware tier recommendations to find models that run on your setup, from consumer GPUs to datacenter clusters.
Open Source LLM Hub
Track the best open-source large language models.
507
Models Tracked
568.2M
Downloads
License:
New Releases
No new releases this week.
Efficiency Rankings
| # | Model | Params | Quality | Efficiency | Price/1M | License |
|---|---|---|---|---|---|---|
| 1 | Qwen3.5 0.8B Alibaba | 1B | 10.5 | 13.13 | $0.02 | apache-2.0 |
| 2 | Qwen3.5 2B Alibaba | 2B | 16.3 | 8.15 | $0.04 | apache-2.0 |
| 3 | Qwen3.5 4B Alibaba | 4B | 27.1 | 6.78 | $0.06 | apache-2.0 |
| 4 | DeepSeek R1 Distill Qwen 1.5B DeepSeek | 2B | 9.1 | 6.07 | N/A | mit |
| 5 | Nanbeige4.1-3B Nanbeige | 3B | 16.1 | 5.37 | N/A | apache-2.0 |
| 6 | Qwen3.5 9B Alibaba | 9B | 32.4 | 3.60 | $0.07 | apache-2.0 |
| 7 | Qwen3 VL 4B Instruct Alibaba | 4B | 9.6 | 2.40 | N/A | apache-2.0 |
| 8 | Qwen3 VL 8B Instruct Alibaba | 8B | 14.3 | 1.79 | $0.18 | apache-2.0 |
| 9 | Qwen3.6 27B Alibaba | 27B | 45.8 | 1.70 | $1.04 | apache-2.0 |
| 10 | Qwen3.5 27B Alibaba | 27B | 42.1 | 1.56 | $0.54 | apache-2.0 |
| 11 | Step3 VL 10B StepFun | 10B | 15.4 | 1.54 | N/A | apache-2.0 |
| 12 | DeepSeek R1 Distill Llama 8B DeepSeek | 8B | 12.1 | 1.51 | N/A | mit |
| 13 | Qwen2.5 Coder 7B Instruct Alibaba | 7B | 10.0 | 1.43 | $0.04 | apache-2.0 |
| 14 | Qwen3.6 35B A3B Alibaba | 35B | 43.5 | 1.24 | $0.36 | apache-2.0 |
| 15 | gpt oss 20b OpenAI | 20B | 24.5 | 1.23 | $0.06 | apache-2.0 |
| 16 | Olmo 3 7B Instruct Allen AI | 7B | 8.2 | 1.17 | $0.13 | apache-2.0 |
| 17 | gemma 4 26B A4B it | 26B | 30.3 | 1.17 | $0.13 | apache-2.0 |
| 18 | DeepSeek R1 Distill Qwen 14B DeepSeek | 14B | 15.8 | 1.13 | N/A | mit |
| 19 | Qwen3.5 35B A3B Alibaba | 35B | 37.1 | 1.06 | $0.35 | apache-2.0 |
| 20 | L EXAONE 4.5 33B LGAI EXAONE | 33B | 30.2 | 0.92 | N/A | other |
| 21 | Qwen3 14B (Non-reasoning) Alibaba | 14B | 12.8 | 0.91 | $0.38 | apache-2.0 |
| 22 | Molmo2 8B Allen AI | 8B | 7.3 | 0.91 | N/A | apache-2.0 |
| 23 | Qwen3 Coder 30B A3B Instruct Alibaba | 30B | 20.0 | 0.67 | $0.12 | apache-2.0 |
| 24 | Qwen3 VL 32B Instruct Alibaba | 32B | 17.2 | 0.54 | $0.18 | apache-2.0 |
| 25 | DeepSeek R1 Distill Qwen 32B DeepSeek | 32B | 17.2 | 0.54 | $0.29 | mit |
| 26 | Qwen3 VL 30B A3B Instruct Alibaba | 30B | 16.1 | 0.54 | $0.23 | apache-2.0 |
| 27 | Qwen3.5 122B A10B Alibaba | 122B | 41.6 | 0.34 | $0.72 | apache-2.0 |
| 28 | gpt oss 120b OpenAI | 120B | 33.3 | 0.28 | $0.07 | apache-2.0 |
| 29 | 80B | 20.1 | 0.25 | $0.34 | apache-2.0 | |
| 30 | Qwen3.5 397B A17B Alibaba | 397B | 45.0 | 0.11 | $0.88 | apache-2.0 |
| 31 | Qwen3 VL 235B A22B Instruct Alibaba | 235B | 20.8 | 0.09 | $0.37 | apache-2.0 |
Best by Hardware
Consumer GPU (16-24GB VRAM)
Models suitable for RTX 4090, M2 Max, and similar hardware.
Prosumer (48-80GB VRAM)
Models for A6000, A100, or multi-consumer-GPU setups.
Data Center (Multi-GPU)
Large models requiring multiple A100/H100 GPUs.