New AI matches rival on hard tests using much less power

New AI matches rival on hard tests using much less power

Reflection AI says its text-only Beam system scores as well as Z.ai's GLM-5.2 on difficult problem-solving tests, while using three to four times less computing power to answer. Beam has 501 billion total settings with 23 billion in use at once, compared with about 744 billion and 40 billion for GLM-5.2, like calling in only needed specialists.

Published

Read at another depth