ARC Prize· @arcprize · X·· 4 小时前AI 评分42
AI 导读
Sol 在高推理模式下 ARC-AGI-3 最佳得分 98.2%,比 Astra 在最高推理模式下的 100% 得分低 1.8 个百分点,但借助提供商适配器测试框架,成本低 64%: - ARC-AGI-3:98.2%,$1.5K(提供商适配器),54.9%,$5.5K(标准) - ARC-AGI-2:85.8%,$0.13/任务 - ARC-AGI-1:97.5%,$0.06/任务 完整结果:arcprize.org/results/openai-g…
正文 · 原文
Sol's best ARC-AGI-3 score of 98.2% at high reasoning was 1.8 percentage points behind Astra's 100% score at max reasoning, but at 64% lower cost with the provider adapter harness:
- ARC-AGI-3: 98.2%, $1.5K (provider adapter), 54.9%, $5.5K (standard)
- ARC-AGI-2: 85.8%, $0.13/task
- ARC-AGI-1: 97.5%, $0.06/task
Full results: arcprize.org/results/openai-g…
来源:ARC Prize · x.com