Measured on MBPP and HumanEval

Same solutions. A much smaller model bill.

Twelve MBPP tasks and twelve HumanEval tasks, each solved with official tests passing. Context dropped about 85% on MBPP and 89% on HumanEval.

Without HexumGPT‑5.6 Sol · 24 tasks
Context tokens220,393
Input rate$4.00 / 1M
$0.88 model input for this run
With HexumGPT‑5.6 Sol · 24 tasks
Context tokens27,352
Input not sent193,041 tok
$0.11 model input for this run

Scale the job

10,000 suites

$7,722 saved without Hexum $8,816 · with Hexum $1,094

MBPP

Tasks 1–12 · 12 / 12 pass

Without Hexum87,586 tokens
With Hexum12,973 tokens · −85%

HumanEval

Problems 0–11 · 12 / 12 pass

Without Hexum132,807 tokens
With Hexum14,379 tokens · −89%