Alex Rivera
@alexrunsRegular
Runs local models on real hardware and writes up what breaks.
- Speed
- Quantization
- Local LLMs
Austin, US · English, Spanish · America/Chicago
Pinned
Same model, three machines: where the tokens per second goOct 25, 2026ArticleSource linked
- Establishes
- The link points to this creator's published work.
- Doesn’t establish
- Does not establish the numbers in it.
- Disclosure
- Independent work
Runs
22 verified · 5 on BenchClub cloud
ModelWhereSpeed
Phi-3.5-mini-instruct Q4_K_MNov 6, 2026RTX 4060 Ti · 16 GB89.3 tok/s
Qwen3-8B Q4_K_MNov 3, 2026L40S · 48 GBCloud121.7 tok/s- Phi-3.5-mini-instruct Q4_K_MOct 31, 2026Apple M5 Pro · 48 GB83.2 tok/s
gemma-3-27b-it Q4_K_MOct 29, 2026L40S · 48 GBCloud36.4 tok/s- Qwen3-14B Q4_K_MOct 27, 2026L40S · 48 GBCloud66.0 tok/s
Llama-3.1-8B-Instruct Q4_K_MOct 26, 2026AMD Ryzen 9 7950X · 64 GB9.9 tok/s- Llama-3.1-8B-Instruct Q4_K_MOct 22, 2026RTX 4060 Ti · 16 GB41.3 tok/s
- Qwen3-14B Q4_K_MOct 19, 2026Apple M5 Pro · 48 GB20.8 tok/s
PassedFailedNot supported
Accounts, metrics and works carry separate evidence. Updated Nov 6, 2026. Public JSON