Ram·6d agoAI Hallucination Rate in 2026: OpenAI Has Models at 3.1% and 23.3%The October 2026 leaderboards run from 1.8% to 77%. In our own test, ChatGPT and Google AI Mode invented 0 of 86 answers.
Lucie Ost·Nov 2What is behind the simplicity of AI nutrition labels?A practitioner’s study notes on ImpactBench
InGoogle Cloud - CommunitybyAmir Davidoff·Nov 1Build your own custom code benchmarkBuild the Eval Factory for Your Software Factory: Benchmark Coding Agents on Your Own Repo with repo2bench (Built on Harbor), and Optimize…A response icon1A response icon1
Orkes Developers·Oct 28How to Benchmark a New LLM on Your Own Tasks: Mistral Large 4 vs GPT-6.1 Sol on ConductorMistral released Large 4 on October 6, 2026 as a public preview. It is a mixture-of-experts model with about one trillion parameters, and…
Joelsstafford·Oct 16I Run d1-ultra Decision Model on PI Trading Benchmark and Failed Badly Losing 17.15%Liquid AI introduced d1-ultra with a bold headline: “Introducing the World’s Best Decision Model.” On PI Trading Arena’s published trading…
Athulramesh·Oct 12LLM Benchmarks, Explained: A Hands-On Look at SWE-benchEvery few weeks a new AI model comes out with a chart full of scores. 92% on this, 78% on that. But what do those numbers mean, and how are…
Munaf·Oct 10AI Is Tested on Questions Nobody AsksMicrosoft researchers found that top models lose 39% of their performance once a task arrives in pieces.
InLevel Up CodingbyAhmed Sozzer·Oct 9A Terminal-Bench task frontier agents could not passI wrote a task for Terminal-Bench, a public benchmark of hard command-line tasks for AI agents. The agent has to build a redactor for…
Inthe/experts Tech TalkbyLars Blom·Oct 9Can you Trust your AI Results?Have you already looked at the general agreeableness of your AI models on their own work? Because you should.
Alan Robinson·Oct 8Benchmarking Apodex: Apodex 1.1 Mini (free on openrouter)Since Oct 1st, there’s a new free model on OpenRouter: Apodex 1.1 Mini. On OpenRouter, new free models often appear with almost no…
