Benchmarks

AI & ML 4 min xAI's Grok 4.7 Leads Two of Seven Benchmarks on Its Own Scorecard Fable 5.1 takes four and GPT-5.6 Sol one. The wins are electrical engineering and a legal agent test, where th... MA Maya Lin 25 Sep Developer Tools 6 min Six of Six on Known-Vulnerable Commits Is a Recall Test Cycode's agentic scanner found every published CVE in a ten-repository benchmark it designed. The false positi... MA Maya Lin 9 Sep AI & ML 6 min The Sub-300ms Latency Figure for GPT-Live Is Not for GPT-Live OpenAI has published an architecture account of its full-duplex voice system and no end-to-end latency benchma... MA Maya Lin 7 Sep AI & ML 7 min Six AI Releases in Three Days, and Two Were New Models The 72-hour launch count folds together two new models, a safeguard variant, two point releases, a pricing tie... MA Maya Lin 4 Sep Developer Tools 6 min One Generated Prompt Beat Up to 63 Per Cent of a Top Conference's LLM Techniques A University of Virginia team reproduced 35 ICSE 2026 techniques and outperformed 37 to 63 per cent of them wi... MA Maya Lin 3 Sep Health & Wellness 7 min The Best Model Called a Death Cap Edible 16 Per Cent of the Time Sixteen models were tested on 1,040 photographs of 55 species. Top-five accuracy is the wrong metric, because... AM Amelia Wong 3 Sep AI & ML 7 min OpenAI's Astra Found Two Zero-Days in a Version Built for Testing OpenAI says Astra found and exploited two zero-days autonomously, in a version modified for testing. That is a... MA Maya Lin 2 Sep AI & ML 7 min Anthropic's AI Beat Human Researchers, but Only on Problems a Computer Can Score Automated agents beat experienced human proposals within six hours, at four dollars an hour against a hundred... KE Kenji Tanaka 30 Aug Cloud & Infra 6 min Nvidia Claims 30x More Work Per Megawatt, Against Its Own Last Generation Vera Rubin is in full production and the efficiency numbers are vendor claims measured against Nvidia's own pr... KE Kenji Tanaka 30 Aug AI & ML 7 min Alibaba's Screen-Operating Agent Scores 82 Per Cent On Phones And 40 On Desktops Qwen-UI-Agent drives real phones, desktops and browsers by looking at the screen rather than calling an API. I... KE Kenji Tanaka 23 Aug