Making AI Headlines · Page 7
Ripple Wins a Coding Claim, Then Meets the Other Benchmarks
· New York time
Ricursive Intelligence is releasing Ripple-65B Preview, a 65-billion-parameter open-weight language model under the Apache 2.0 license. The company says it leads open-source US models on the TerminalBench 2.1 coding benchmark and puts its total research and development cost at $32 million. The claim has specific boundaries: it concerns a particular benchmark and a particular group of models. The Decoder reports that Ripple falls behind the newly released western open model Inkling 2 Small on other benchmarks. That does not erase the company's coding result, but it does make the contest more interesting than a single universal ranking. The open license gives developers broad permission to use, modify and redistribute the model. Open weights also let them run and test it themselves, subject to the computing resources required. For users choosing a model, the relevant result will depend on the tasks they actually need to perform as well as the cost of doing them. Benchmark victories are the AI industry's favorite way to print a medal before the race has finished. Ripple's medal comes with a clear inscription: TerminalBench 2.1, among open-source US models, according to its maker. That is a useful achievement to evaluate, and a much smaller object than the trophy labeled 'best at everything' that every model seems to want on its desk.
Read this story in the full newspaper →Satire. Not factual reporting. Read the original reporting from the linked publishers.