Poetiq’s Meta-System Automatically Builds a Model-Agnostic Harness That Improved Every LLM Tested on LiveCodeBench Pro Without Fine-Tuning

Poetiq says its Meta-System automatically built and optimized a model-agnostic coding harness for LiveCodeBench Pro.

The system reportedly boosted scores for models including GPT 5.5 High, Gemini 3.1 Pro, and Gemini 3 Deep Think.

Poetiq says the gains came without fine-tuning or access to model internals, only through inference orchestration.

If validated, the result suggests automated harnesses could improve coding benchmark performance across many LLMs without changing the models.