The Qxern-v6 system enables two LLMs to communicate via 32 compressed latent tokens while preserving exact symbol accuracy through an adaptive deterministic Abstract Syntax Tree (AST) sidecar. Built by a 15-year-old developer, the architecture uses Qwen2.5-Coder-1.5B to compress code and a frozen Qwen3.5-0.8B decoder to interpret it without seeing raw text.

  • Function name accuracy reaches 0.87, recovering from 0.00 in pure latent models.
  • Parameter count accuracy hits 0.93, outperforming the Text Relay baseline by +0.30.
  • Median latency drops to 511 ms, achieving a 2.4× speedup over the 1222 ms text relay.
  • The system combines semantic latents with ~30 tokens of deterministic AST data via an adaptive router.

This hybrid approach demonstrates that continuous latents can efficiently carry code semantics at lower latency while the sidecar restores necessary symbolic precision without retraining the decoder.