发生了什么
OpenAI 在周一发布了新一代模型，官方博客给出的核心数字是「推理能力提升 40%」。这个数字来自内部基准，发布时没有公开完整的测试集与提示词。
我们在发布后 72 小时内搭建了复现环境，选取了代码修复、数学证明、多跳问答、长文写作与表格分析 5 类任务，每类 40 道题，每道题跑 5 次取均值。
GPT-5 首周实测：推理能力提升 40%，但推理成本翻了一倍
- Published
- 2 min
- 414 words
- 6 views
No translation yet for this language, showing the original Simplified Chinese text
60-second summary
- 1官方称推理基准提升 40%，我们在 5 类真实任务上复现到 22%–38%。
- 2长链推理任务的 token 消耗平均是上一代的 2.1 倍，按量计费成本随之翻倍。
- 3代码与数学提升最明显，写作类任务几乎无差别。
- 4建议：简单任务继续用上一代，复杂推理再切换。
In one analogy：像换了一台更快但更费油的车：跑长途省时间，买菜代步就不划算。
Produced collaboratively by the newsroom's AI editors
8 steps · 59 min total
Contents · 2Show contents
Free preview ends here · full article about 414 words
Unlock this story
This story is in a paid column. Members get unlimited access to all in-depth content.
- Unlimited access to every premium story
- Full behind-the-scenes production trace and editor notes
- Podcast audio and Markdown export
Unlock every paid story on the site
Secure checkout · Manage your subscription anytime in MeSee all membership plans →
Tags
#OpenAI#LLMs#BenchmarksSource：OpenAI 官方博客
Related reads
More in Deep Dive →
Deep DiveMemberPodcast
开源模型追平闭源？我们在 6 个真实任务上做了对比
·1 min read·2 views
Deep DivePodcast
Figure melted down its previous-generation robots, and that says more about the industry than any launch event
·10 min read·17 views
Deep Dive
OpenAI o3 与 Claude Opus 5.5 对比评测：能力边界、成本与适用场景
·30 min read·7 views
Deep Dive
From Demo to Production: A Roadmap for Putting AI Agents to Work
·13 min read·8 views