The Kohnex Blog

Proof over promises: benchmarks, reasoning, and real results.

BENCHMARK RESULTS ·

DeepSeek V4 Flash + Kohnex: +22.5 Points on FORGE-100

Same model, same max effort — 42.7% to 65.2% across 100 custom expert tasks, with the full category breakdown.

Read more →
MANIFESTO ·

Proof Over Promises: Why We Publish Our Benchmarks

Our evaluation policy in writing — full breakdowns, unflattering numbers included, methodology before marketing.

Read more →
REASONING ·

Small Models Think Big: What MiniCPM5-2B Proves

A 2.5B model beating giants on math and code confirms it — reasoning beats raw scale.

Read more →