Open models and the hardware they need
Memory needed by popular open-weight models at 4, 8 and 16 bits, and the GPUs, Macs and mini PCs that can run them.
| Model | Parameters | 4-bit | 8-bit | 16-bit | Max context | Released |
|---|---|---|---|---|---|---|
| Qwen3 1.7BAlibaba (Qwen) | 2B | 1.1 GB | 1.8 GB | 3.5 GB | 40K | April 2025 |
| Llama 3.2 3BMeta | 3.2B | 2 GB | 3.4 GB | 6.4 GB | 128K | September 2024 |
| Llama 3.1 8BMeta | 8B | 4.9 GB | 8.5 GB | 16 GB | 128K | July 2024 |
| Gemma 4 E4BGoogle | 8BActive: 4B | 5 GB | 8.2 GB | 16 GB | 128K | March 2026 |
| Qwen3 8BAlibaba (Qwen) | 8.2B | 5 GB | 8.7 GB | 16 GB | 40K | April 2025 |
| Qwen3.5 9BAlibaba (Qwen) | 9.7B | 5.7 GB | 9.5 GB | 19 GB | 256K | February 2026 |
| Gemma 4 12BGoogle | 12B | 7.1 GB | 13 GB | 24 GB | 256K | May 2026 |
| gpt-oss-20bOpenAI | 21BActive: 3.6B | 12 GB | 12 GB | 14 GB | 128K | August 2025 |
| Mistral Small 3.2 24BMistral AI | 24B | 14 GB | 25 GB | 48 GB | 128K | June 2025 |
| Qwen3.6 27BAlibaba (Qwen) | 28B | 17 GB | 29 GB | 56 GB | 256K | April 2026 |
| Gemma 4 26B A4BGoogle | 26BActive: 4B | 17 GB | 27 GB | 52 GB | 256K | March 2026 |
| Gemma 4 31BGoogle | 31B | 18 GB | 33 GB | 63 GB | 256K | March 2026 |
| Qwen3-Coder 30B A3BAlibaba (Qwen) | 31BActive: 3.3B | 19 GB | 32 GB | 61 GB | 256K | July 2025 |
| Qwen3 32BAlibaba (Qwen) | 33B | 20 GB | 35 GB | 66 GB | 40K | April 2025 |
| Qwen3.6 35B A3BAlibaba (Qwen) | 36BActive: 3B | 22 GB | 37 GB | 72 GB | 256K | April 2026 |
| Llama 3.3 70BMeta | 71B | 43 GB | 75 GB | 141 GB | 128K | December 2024 |
| gpt-oss-120bOpenAI | 117BActive: 5.1B | 63 GB | 63 GB | 65 GB | 128K | August 2025 |
| Qwen3.5 122B A10BAlibaba (Qwen) | 125BActive: 10B | 77 GB | 130 GB | 250 GB | 256K | February 2026 |
| Qwen3 235B A22B (2507)Alibaba (Qwen) | 235BActive: 22B | 142 GB | 250 GB | 470 GB | 256K | July 2025 |
| Qwen3.5 397B A17BAlibaba (Qwen) | 403BActive: 17B | 244 GB | 422 GB | 807 GB | 256K | February 2026 |
| DeepSeek R1 (0528)DeepSeek | 671BActive: 37B | 405 GB | 713 GB | 1,342 GB | 160K | May 2025 |