Researchers replicated reasoning evaluations from prior work on Opus 4.5, then tested three newer models—Claude Fable 5, Opus 5, and GPT-5.6-Sol—using single-forward-pass evaluation methods. Fable 5 achieved 87.6% accuracy on Gen-Arithmetic with 10 problem repeats (compared to previous best around 60%), and GPT-5.6-Sol showed substantial uplift from filler tokens and problem repeats across datasets, with performance roughly doubling on the 3-hop task.
Summaries like this, in your inbox every morning.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Morgan Stanley is lending capital to the AI build-out and using AI internally, while CEO Ted Pick told analyst…
At RCR Wireless' 6G Forum, Verizon CTO Yago Tenorio said AI could push the downlink-to-uplink ratio from 9:1 t…
John Deere is testing the JD AI assistant in its Operations Center, according to Farmtario
Verizon rolled out complimentary perks for subscribers, including Verizon Digital Secure, Verizon Family with…
Booking Holdings launched Vayo, its second AI-native startup, in private beta with a waitlist at vayo.com, and…
A blog ran a Gemini simulation of all 30 teams entered in the October 18 Princess Ekiden, picking Daiso, Panas…
