跳到正文
原文
X:ARC Prize (@arcprize)· @arcprize·· 2 天前AI 评分42

ARC Prize 回应 Qwen3.8-27B 榜单成本估算方式,ARC-AGI-3 结果待测试完成

R to @Alibaba_Qwen: Leaderboard costs were estimated by applying Alibaba Cloud's listed token prices to recorded token usage, rather than using our dedicated-inference charges. We will share ARC-AGI-3 results once testing is complete. Full results: https://arcprize.org/results/alibaba-qwen3-8-27b

AI 导读

ARC Prize 表示 Qwen3.8-27B 的榜单成本是按阿里云公开 token 价格乘以记录的 token 用量估算,而非其专用推理收费。该模型经 Baseten 测试,使用自带 chat template:low 要求思考简短聚焦,xhigh 要求仔细思考并检查假设,medium 两者都不加。ARC-AGI-3 结果将在测试完成后公布。

来源:X:ARC Prize (@arcprize) · x.lingyaoai.com