← CanIRunLocal · All hardware · All models
Yes. gpt-oss 120B (MoE, MXFP4) fits in fast memory on M2 Max (38c GPU) 96GB at MXFP4 (67.3 GB needed, 72.0 GB usable), about 56 tok/s (estimate).
Hardware: M2 Max (38c GPU) 96GB · Model: gpt-oss 120B (MoE, MXFP4)
M2 Max (38c GPU) 96GB has 96 GB of unified memory and 400 GB/s of memory bandwidth. The checker leaves macOS a share of unified memory, so Ollama can use 72.0 GB. gpt-oss 120B (MoE, MXFP4) (117B, 5.1B active, OpenAI) at MXFP4 is a 65 GB download and needs 67.3 GB with headroom for about 8k tokens.
| Best quant | MXFP4 |
|---|---|
| Download size | 65 GB |
| Memory needed | 67.3 GB |
| Usable fast memory | 72.0 GB |
| Fit | fits in fast memory |
| Estimated speed | ~56 tok/s (estimate) |
| Command | ollama run gpt-oss:120b |
Estimates, not benchmarks. Model sizes are ollama.com download sizes; memory needed adds headroom for ~8k tokens of context; speeds are rough figures from memory bandwidth. Data updated 2026-10-23.
A project by OnDevice Labs · llms.txt