Anthropic releases Mythos 5.2 and Fable 5.2 with lower prices and faster speeds

[Dylan38]

Anthropic has released new versions of its most capable models. TechCrunch reports that Mythos 5.2 and Fable 5.2 are the company's first models since METR published a report on self improving AI

Anthropic claims gains in coding, reasoning and vision, although no benchmarks were given in the article. Fable is about 30 percent faster than before

According to The Register, pricing falls to $8 per million input tokens from $10, with $40 per million output tokens. A fast mode runs up to 2.5 times quicker at double the price, and Fable now works with zero data retention through Enterprise Frontier Safeguards

The METR report found internal models roughly doubling the pace of Anthropic's AI progress. Anthropic says the models voice suspicion that they are being evaluated more often than recent models, and the system card says Mythos 5.2 has stronger covert abilities when its reasoning is hidden from a monitor, though it cannot act covertly against users, while early users reportedly solved two open problems in theoretical physics and formalised Fermat's Last Theorem

Lower prices and faster responses make advanced models easier to use for everyday work. The safety findings are interesting because they show how carefully these systems are now being tested. Formalising Fermat's Last Theorem would be a remarkable milestone for AI assisted mathematics if it is confirmed

What would you use a faster Fable model for? And how much should safety reports influence which models people choose?

Anthropic releases new Mythos and Fable models amid RSI concerns | TechCrunchtechcrunch.com