Jana Nováková
@jana-novakRegular
PhD student. Reproduces other people's benchmark claims.
- Reproducibility
- Evaluation
Prague, Czechia · English, Czech · Europe/Prague
Pinned
I re-ran 12 viral benchmark posts. Here's what held up.Oct 26, 2026ArticleIndependently reproduced
- Establishes
- Another member re-ran it and matched the reported results.
- Doesn’t establish
- Does not establish results on other hardware.
- Disclosure
- Independent work
Runs
12 verified · 4 on BenchClub cloud
ModelWhereSpeed
Llama-3.1-8B-Instruct Q4_K_MNov 4, 2026Apple M1 Max · 64 GB52.4 tok/s
gemma-3-12b-it Q4_K_MNov 2, 2026L40S · 48 GBCloud81.7 tok/s
Phi-3.5-mini-instruct Q4_K_MOct 28, 2026Apple M1 Max · 64 GB112.0 tok/s
GLM-4.7-Flash Q4_K_MOct 24, 2026L40S · 48 GBCloud201.7 tok/s
Qwen3-14B Q4_K_MOct 20, 2026Apple M1 Max · 64 GB28.2 tok/s
gpt-oss-20b MXFP4Oct 13, 2026Apple M1 Max · 64 GB68.3 tok/s- Qwen3.5-2B Q4_K_MOct 6, 2026Apple M1 Max · 64 GB132.0 tok/s
- gpt-oss-20b MXFP4Oct 6, 2026L40S · 48 GBCloud173.8 tok/s
PassedFailedNot supported
Accounts, metrics and works carry separate evidence. Updated Nov 3, 2026. Public JSON