Benchmark evidence

Verified external-baseline results for FlowMemory FlowMemory. See the methodology page for the complete test design and claim boundaries.

LLAMA.CPP

50.31× median

Final hot-prefix request across three paired runs: 48.36×, 50.31×, and 56.39×.

OLLAMA

20.76× median

End-to-end final request across three paired runs: 24.24×, 20.76×, and 17.49×.

CORRECTNESS

Outputs identical

Every generated output matched its external baseline in the published suites.

Review methodology