Signal 65 test: Astra completes business tasks more reliably and cheaply
At its highest reasoning setting, Astra finished 279 of 280 tasks at $1.51 per successful task, about 39% less than Fable 5.1, and made up no answers.
OpenAI’s GPT-6 Astra completed business tasks more reliably and at a lower cost than Claude Fable 5.1 in testing by @Signal_65. At its highest reasoning setting, Astra: • Completed 279 of 280 jobs from start to finish. • Cost $1.51 per successful task vs. Claude’s $2.46, about 39% less. • Made up no answers on questions that couldn’t be answered from the provided information. Astra’s cheaper medium setting cost $0.94 per successful task, but made up answers on 2.3% of those unanswerable questions. Note: Its PINNACLE benchmark tests whether AI can finish multi-step business jobs and avoid making up answers when information is missing.
This post quotes another post: x.com/signal_65/status/2096293645783576658