benchmarks - Abrarqasim Blogs

Efficiency Hallucination: Every Model Rewrote Code That Couldn’t Get Faster

A Columbia paper found LLMs edit already-optimal code 100% of the time. One prompt line halves it with zero false refusals. I reproduced it and…

Rayyan | September 18, 2026 | 9 min

AI

LLM Benchmarks: The API Scores Better Than the App You Actually Use

A Stanford paper finds API benchmark scores run 3.4 points higher than the same models in their chat apps. What that means for how I…

Rayyan | September 12, 2026 | 8 min

AI

ARC-AGI-3 and GPT-6 Astra: Two Scores, One Model, and the Harness Gap

GPT-6 Astra scored 62.7% and 99.9% on ARC-AGI-3 with the same weights. The harness made the difference, and that should change how you evaluate your…

Rayyan | September 6, 2026 | 9 min

AI

LLM Evaluation Tools Keep Missing the Long Game

My LLM evaluation tools said the agent was fine. Two long-horizon benchmarks explain why it fell apart in week three, and what I test differently…

Rayyan | August 10, 2026 | 7 min

AI

Your reasoning model is more fragile than the benchmarks say

A new paper tested 8 reasoning models with 14 formatting perturbations. Open-weight models lost up to 55% accuracy. Here's what that means for production use.

Rayyan | April 15, 2026 | 6 min