Models · chat and writing, reasoning and maths, coding
gpt-oss 120B: VRAM requirements and GPUs that run it
OpenAI's large MoE. 5.1B active, so fast once it fits. 80 GB class or a big Mac. At Q4 with 8K context it needs about 77 GB, so you want a GPU with at least 96 GB of VRAM or a Mac with 96 GB of unified memory.
Memory needed by quantization
| Quant | Weights | Total at 8K | Total at 32K |
|---|---|---|---|
| Q4_K_M | 76 GB | 77 GB | 79 GB |
| Q5_K_M | 91 GB | 92 GB | 94 GB |
| Q8_0 | 136 GB | 137 GB | 139 GB |
| FP16 | 257 GB | 258 GB | 260 GB |
Totals include about 10% runtime overhead and the KV cache. Longer conversations cost real memory: going from 8K to 32K context adds about 2.3 GB at any quantization. It is a mixture-of-experts model: 117B parameters in total but only about 5.1B are used for each token, so it runs much faster than a dense model of the same size and stays usable even when part of it sits in system RAM. Most people should download Q4_K_M; Q5 or Q8 only make sense if you have spare memory. Read which quantization to pick.
Which GPUs can run gpt-oss 120B?
No single consumer GPU we track runs it fully in VRAM at 8K context.
Running it on a Mac
Apple Silicon Macs can give a model about 72% of their unified memory. gpt-oss 120B fits from 96 GB (Apple M3 Max, about 30 tokens per second). With a faster chip like the Apple M3 Ultra (96 GB) expect around 61 tokens per second. Compare the options in our Mac vs NVIDIA guide.
Similar size models: Llama 3.3 70B, DeepSeek-R1 Distill 70B, Llama 4 Scout 109B-A17B.
Compare gpt-oss 120B
Questions
How much VRAM does gpt-oss 120B need?
About 77 GB at Q4_K_M with 8K context, 137 GB at Q8_0, and 258 GB at full FP16. In practice that means a card with at least 96 GB of VRAM for Q4.
Can gpt-oss 120B run on an 8 GB GPU?
No. It needs about 77 GB at Q4, well beyond 8 GB even with offload.
Can I run gpt-oss 120B on a Mac?
Yes. The smallest configuration that runs it fully is Apple M3 Max with 96 GB of unified memory, at about 30 tokens per second. More memory bandwidth (Pro, Max, Ultra chips) means faster replies.
Estimates based on model size, architecture and memory bandwidth. Real numbers vary with software, drivers and context length.