Written by Jakub Rusinowski · Last updated July 21, 2026
Best all-round pick: GPT-oss 120B
The NVIDIA A100 80GB has 80 GB of VRAM, of which about 80 GB is available to a model. The largest model it can hold is Devstral-2 123B (123B, 75.1 GB at Q4_K_M).
| Model | Score | Memory at Q4_K_M | Context | Licence |
|---|---|---|---|---|
| GPT-oss 120B | 96.4 | 71.3 GB | 128K | Apache-2.0 |
| Qwen 3.5 122B-A10B (MoE) | 94.4 | 74.5 GB | 125K | Apache 2.0 |
| GLM-5.1 72B | 94.1 | 44.3 GB | 125K | MIT |
| Qwen 3.5 72B | 93.7 | 44.3 GB | 125K | Apache 2.0 |
| Llama 3.3 70B Instruct | 92.9 | 43.1 GB | 128K | Llama Community |
| Qwen 3.5 122B-A10B | 92.8 | 74.5 GB | 128K | Apache-2.0 |
| Workload | Recommended on this GPU |
|---|---|
| Coding | Devstral-2 123B (99.1) Qwen3-Coder 80B-A3B (MoE) (98.6) Qwen 3.5 122B-A10B (MoE) (96.8) |
| General assistant | GPT-oss 120B (96.4) Qwen 3.5 122B-A10B (MoE) (94.4) GLM-5.1 72B (94.1) |
| Reasoning | GLM-5.1 72B (100) Qwen 3.5 122B-A10B (MoE) (97.1) Gemma 4 27B ⭐ (96.8) |
| RAG | Llama 4 Scout 17B (94.4) Gemma 4 31B (94.1) Llama 4.5 Scout (93.1) |
| Agents | Devstral-2 123B (95.3) Ternary Bonsai 27B (90.1) Nemotron-Cascade 2 30B-A3B (89.4) |
| Vision | Qwen 3.5 72B (99.2) Llama 3.2 90B Vision Instruct (97.1) Qwen 3.7 35B-A3B (94.5) |
GPT-oss 120B is the strongest all-round pick that fits its 80 GB.
Devstral-2 123B — 123B parameters, needing 75.1 GB at Q4_K_M.
About 80 GB of its 80 GB, before the desktop and runtime overhead are accounted for.