That means that Google’s technology—which consumers can tap into via their Google Home appliances or Google Assistant on mobile devices—gets 19 out of 20 words right. The news came out of the Google I/O tech conference on Wednesday.
Get Data Sheet, Fortune’s technology newsletter.
Speech recognition software—the engine that enables Apple (aapl) Siri, Google (goog) Assistant, and Amazon Alexa to “understand” what people say—is a major front in the larger battleground of artificial intelligence.
And all of the big vendors—except for one—have taken turns claiming victory in terms of error rates.
In March, IBM (ibm) researchers claimed a breakthrough word error rate of 5.5% for that company’s technology. In October, Microsoft’s speech brain trust said they’d wrestled its error rate down to 5.9% from 6.3% the previous month.
Natural language processing (NLP) is obviously a hotly contested battleground. Big vendors such as Amazon (amzn), which is pushing its Alexa speech recognition in the home and increasingly in businesses, want consumers and businesses to use their speech technology to interact with software—while they also collect lots of data about those users in the process.
Related: Stop Screaming at Siri Right Now
Generally, researchers say that people miss one or two words out of every 20 spoken, so the technology is getting pretty good, but not good enough. In December, Adam Coates, the head of Chinese search giant Baidu’s AI lab in Sunnyvale, Calif., told Bloomberg that his goal is to get that error rate down to a meager 1% because even 95% accuracy won’t cut it. Anyone who’s had a “conversation” with Siri or Alexa or Microsoft Cortana go off the rails can understand that.
Expect more leap frogging claims to come.