Give opus 5 a chart with no tools and it reads it wrong 70% of the time

Almost nobody benchmarking what happens when you hand it a tool numbers that actually matter for design work: Chart reading [ Chartography ]: 29.6% with no tools > 83.0% with a basic image cropping tool

YARDSep 121

Artificial intelligence no longer just learns from humans

It can also contribute to the development of other artificial intelligence models. 🤖⚡ In Anthropic's experiment, Claude Sonnet 5 was used to mitigate security issues in the early release of the more powerful Claude Opus 4.8. sonnet

beeteknoSep 1

Claude Opus 5, GPT-5.6 Sol, and Gemini 3.7 Flash

Fede(URU) 🇺🇾Sep 1

This fight is not even close and everyone picked the wrong side

Moon dev: "i'm the biggest claude guy alive and even i have to admit what just happened" "opus 5 versus gpt 5.6 went one versus two and the math completely destroyed me"

stacksbyraySep 1