# GPT-6 Astra 在 WeirdML v3 上大幅领先 Fable 5.1

- 来源：Haider. (@haider1)
- 发布时间：2026-09-18 21:00
- AIHOT 分数：36
- AIHOT 链接：https://aihot.news/items/cmu6ywcfe0jyerowkert4dtw6
- 原文链接：https://x.com/haider1/status/2100932864703656296

## AI 摘要

GPT-6 Astra 在完全智能体化的 ML 基准 WeirdML v3 上以 42% 得分、$27 成本、153k 输出 token 击败 Fable 5.1 的 26%、$32、251k 输出 token，效率差距明显。

## 正文

GPT-6 Astra is crushing Fable 5.1 on the fully agentic ML benchmark WeirdML v3

astra: 42% — $27 — 153k output tokens fable 5.1: 26% — $32 — 251k output tokens

pretty absurd efficiency gap

but i think the more worrying part is how badly open-weight models are still lagging

### 引用推文

> Håvard Ihle：Introducing WeirdML v3, a fully agentic benchmark featuring 11 complex hand-made tasks. Models must explore and understand unfamiliar data, develop ML and data ...