Anthropic的新模型Sonnet 5，基准测试上超过自家大哥Opus 4.8，价格还更低，挺能打的。
Anthropic推出Claude Sonnet 5，在GDPval-AA v2知识工作测试中得分1,618，超越前代Sonnet 4.6和更贵的Opus 4.8。该模型在网络安全任务上得分远低于美国政府当前封锁的模型。Sonnet 5进一步缩小了与Opus系列的价格-性能差距。
Anthropic's new Claude Sonnet 5 closes the gap to the pricier Opus model series
Anthropic released Claude Sonnet 5, which beats its predecessor Sonnet 4.6 across all benchmarks and even edges past the larger Opus 4.8 on the GDPval-AA v2 knowledge work test with a score of 1,618. Anthropic is also quick to point out that the model scores far below the models the US government currently has blocked when it comes to cybersecurity tasks, a likely deliberate signal given the ongoing debate. The article Anthropic's new Claude Sonnet 5 closes the gap to the pricier Opus model series appeared first on The Decoder .