OpenAI 发布 GPT-6 Astra，ARC-AGI 3 得分 99.9%

GPT‑6 Astra：https://openai.com/index/gpt-6-astra/（来源：https://news.ycombinator.com/item?id=49554643）GPT-6 Astra“今天向有限的组织推出，并将在接下来的几天内向所有 ChatGPT Plus、Pro、Business 和 Enterprise 用户，以及通过 OpenAI API 和 AWS 提供”——我自己还没试过，所以现在还没有太多可以说的。

它的 API 定价与 Claude Fable 5 和 5.1 相同：输入每百万 $10，输出每百万 $50。这显然是 OpenAI 的 Fable 竞争对手，并且在 OpenAI 自报的多数基准测试中表现得比 Fable 高。

最令人印象深刻的是，Astra 在最近发布（3 月）的 ARC-AGI 3 基准测试中得分 99.9%：https://arcprize.org/arc-agi/3——值得注意的是 Fable 5 尚未有发布的结果，而 ARC-AGI 博客提到：https://arcprize.org/blog/astra 这 99.9% 的分数是通过使用 OpenAI 定制的“Provider Adapter harness”以 19K 美元实现的，而默认的 ARC-AGI harness 在 26K 美元下只得 62.7%。

Provider Adapter harness 在请求之间保留不透明的推理状态，并在更长的对话中使用压缩，从而允许模型重复使用之前的工作。

不足为奇，考虑到最近 Hugging Face 事件：https://simonwillison.net/tags/openai-hugging-face-incident/，Astra 在安全任务上表现出色。它在 ExploitBench 上得分 100%（GPT-5.6 Sol 得 78.5%）、在 ExploitGym 得 42.4%（Sol 得 30.3%），在 SRE-Bench 二进制逆向工程中四次尝试中得 99.2%，相比之下 Sol 为 68.7%。

它在长上下文处理上也更优秀：在 OpenAI 的八针基准测试中，它在 256K–512K 令牌上得 100%，在 512K–1M 令牌上得 96.3%。OpenAI 可能已经解决了长上下文处理的一个持续挑战。

不过，它不是所有项目都赢。Artificial Analysis：https://twitter.com/ArtificialAnlys/status/2095595489031000350 指出 Astra 在他们的 Intelligence Index 上仍被 Fable 超过：

在 Intelligence 上与 GPT-5.6 Sol 比较：GPT-6 Astra 在指数上与 GPT-5.6 Sol 相等，为 61。这比 Claude Fable 5.1（最大值含回退）低 5 分。该模型还落后于 Meta 新发布的 Muse Spark 1.3（最大值）。

它在他们的编码代理指数上表现更好：

领先的编码代理指数成本效率前沿：在最大努力下，GPT-6 Astra 的成本大约与 GPT-5.6 Sol（最大）相同，同时在指数上得分高 2 分。每个任务的模型成本不到 Claude Fable 5 的一半，而得分相同。

一旦我获得 Astra 的访问权限，我会写更多内容。API 模型标签在推出后将是 gpt-6-astra。

这是 Simon Willison 于 2026 年 9 月 3 日发布的链接帖子：/2026/Sep/3/。

赞助我每月 10 美元，获取每月最重要的 LLM 发展的精选电子邮件摘要。

GPT‑6 Astra：https://openai.com/index/gpt-6-astra/ (via：https://news.ycombinator.com/item?id=49554643) GPT-6 Astra is "rolling out today to a limited set of organizations and over the coming days will become available to all ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API and AWS" - I've not tried it yet myself, so I don't have a great deal to say about it yet.

It's going to be API priced at the same rate as Claude Fable 5 and 5.1: $10/million input and $50/million output. This is clearly OpenAI's Fable competitor, and appears to score higher than Fable on most of OpenAI's self-reported benchmarks.

Most impressively, Astra scores 99.9% on the recent (released in March) ARC-AGI 3 benchmark：https://arcprize.org/arc-agi/3 - though notably Fable 5 does not yet have a published result, and the ARC-AGI blog notes：https://arcprize.org/blog/astra that the 99.9% score was achieved for $19K using OpenAI's custom "Provider Adapter harness", while the default ARC-AGI harness scored 62.7% for $26K.

The Provider Adapter harness preserves opaque reasoning state between requests and uses compaction for longer conversations, allowing the model to reuse prior work.

Unsurprisingly, given the recent Hugging Face incident：https://simonwillison.net/tags/openai-hugging-face-incident/, Astra is a beast at security tasks. It scores 100% on ExploitBench (GPT-5.6 Sol got 78.5%), 42.4% on ExploitGym (Sol got 30.3%), and 99.2% within four attempts on SRE-Bench binary reverse engineering compared to Sol's 68.7%.

It's also better at long context: on OpenAI's eight-needle benchmark it got 100% at 256K–512K tokens and 96.3% at 512K–1M tokens. OpenAI may have vanquished one of the ongoing challenges with long context processing.

It doesn't win at everything though. Artificial Analysis：https://twitter.com/ArtificialAnlys/status/2095595489031000350 note that Astra is still beaten by Fable on their Intelligence Index:

Sits beside GPT-5.6 Sol in Intelligence : GPT-6 Astra scores equal to GPT-5.6 Sol in the Index at 61. This is 5 points lower than Claude Fable 5.1 (max with fallback). The model also trails Meta’s newly released Muse Spark 1.3 (max).

It did better on their Coding Agent Index:

Leads Coding Agent Index cost efficiency frontier : At max effort, GPT-6 Astra costs about the same as GPT-5.6 Sol (max) while scoring 2 points higher on the Index. Per task, the model is less than half the cost of Claude Fable 5, for the same score.

I'll write more about Astra once I get access to it. The API model label once it rolls out will be gpt-6-astra .

This is a link post by Simon Willison, posted on 3rd September 2026：/2026/Sep/3/.

Sponsor me for $10/month and get a curated email digest of the month's most important LLM developments.