webAI released TwIL-8B, an 8.8B reasoning model post-trained from IBM Granite 4.2 8B. It posts 98.2 on MATH-500 and 81.6 on LiveCodeBench, but trails similar-size models on knowledge benchmarks. The post webAI Releases…
新模型会重置能力与性价比的前沿。每次发布改变「每美元能做什么」,团队就要重新评估该基于哪个模型构建。
摘要仅供参考,请点击来源链接查看全文。演示条目为示意。