Recursive Self-Improvement Delivers New SOTA Coding Performance
SMRTR summary
Poetiq's Meta-System automatically builds and optimizes AI inference harnesses — no fine-tuning required — and just hit a new state-of-the-art on LiveCodeBench Pro. Using only standard API access, it boosted Gemini 3.1 Pro by 12.3% and pushed GPT 5.5 to 93.9%, surpassing all tested models. The harness is model-agnostic, meaning gains transfer across providers and model sizes.
SMRTR provides this summary for quick context. The original article belongs to Hacker News.
Read the original article