Latest

Key Takeaways - Qwen3.6-27B performs impressively on coding benchmarks, showing competitive results against larger models. And it's a 27B model, not a 397B monster. - The Q4_K_M quantized version has manageable hardware requirements. Your RTX 4080 handles it. A used RTX 3090 works too. -

Key Takeaways - Roberto Serrano, Brown University economist, flagged at least 50 students using ChatGPT on the midterm - Midterm: 96/100 average with 40 perfect scores out of 89. Final (in-person): 48/100 average - 22 of 27 students who skipped the final had aced the midterm with a

TL;DR - GLM-5.2 scored 77.8 on SWE-bench Verified and 56.2 on Terminal Bench 2.0, surpassing every open-weight model to date and beating GPT-5.5 on multiple long-horizon coding benchmarks. - The model costs roughly $1.40 per million input tokens and $4.40 per million

Key Takeaways - Ford's bet-everything-on-AI quality control strategy imploded. Billions in recalls. Defects everywhere. Course correction took three years. - The company quietly rehired 350+ veteran engineers it had replaced. Bloomberg called them "gray beards." - Ford's own VP said it straight: they "

TL;DR - Doubleword.ai ran all benchmarks from Artificial Analysis. Average lag between open-weight and closed-source LLMs sits at a flat level. Hasn't moved. - Coding's the outlier: open models improved significantly on that specific index. - Everything else is getting worse, not better. Some

Three hundred a month for Grok access felt steep until I did the math on AI infrastructure at scale. That number stuck in my head when OpenAI dropped GPT-5.6 Sol's benchmarks this week. 750 tokens per second. Not a research preview. Not a theoretical paper. Something you