← CanIRunLocal · All hardware · All models
Reflection · 501B parameters (23B active) · chat, reasoning, coding.
Ollama lists beam:cloud as a cloud model, so there is no local download. The fits below use verified Hugging Face weights.
Yes on 1 machine in this list. The smallest is 4x Mac Studio M5 Ultra (80c GPU) 256GB: Q8_0, about 535.1 GB, roughly 15 tokens per second, via llama.cpp. llama-server -hf unsloth/Beam-1.0-501B-A23B-GGUF:Q8_0 --rpc node2:50052,node3:50052,node4:50052 -c 8192 # run rpc-server on each other node
llama-server -hf unsloth/Beam-1.0-501B-A23B-GGUF:Q8_0 --rpc node2:50052,node3:50052,node4:50052 -c 8192 # run rpc-server on each other node| Hardware | Runtime | Quant | Memory | Fit | Speed | Command |
|---|---|---|---|---|---|---|
| 4x Mac Studio M5 Ultra (80c GPU) 256GB | llama.cpp | Q8_0 GGUF | 535.1 GB | fits in fast memory | ~15 tok/s | llama-server -hf unsloth/Beam-1.0-501B-A23B-GGUF:Q8_0 --rpc node2:50052,node3:50052,node4:50052 -c 8192 # run rpc-server on each other node |
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 2026-10-19.
A project by OnDevice Labs · llms.txt