benchmarks
4 stories in this blend
Superintelligence·5d agoBenchmark Report Indicates Gap Between American and Chinese AI Performance Is Narrowing
A Bloomberg Intelligence evaluation shows top Chinese artificial intelligence models trailing leading American systems by just three percent on standardized benchmarks. The shift is largely driven by DeepSeek's recent releases and improved optimization for domestic silicon.
Superintelligence·5d agoCrowdsourced Design Rankings Reveal Strong Performance From Chinese Web Models
User voting data from Design Arena ranks model outputs on front end and 3D web design tasks based on direct head to head user preferences. Chinese models secured multiple top ten spots, showing competitive strength in practical interface generation.
Superintelligence·Oct 2Frontier AI Benchmarks Outperform Human Certified Accountants on Month-End Close Tasks
Testing conducted by Mercor revealed that top AI systems achieved near perfect scores on complex accounting tasks where licensed professionals averaged below forty percent. The evaluation highlighted rapid model improvement over the past year, though researchers note human accountants perform broader workplace roles.
There's An AI For That·Aug 26OpenAI releases benchmark scores for its proprietary hardware
Initial performance tests on OpenAI custom processing chips demonstrate lower latency and nearly double the energy efficiency compared to standard silicon. Earlier models were utilized to design the new hardware that now runs modern systems.
