打架成本警示牌
Anthropic新模型偷吃瓜！最强Fable 5爆冷_我的网站
A |
Zhang Zhiheng, a boy from east China's Shandong Province, was the landmark visitor. He visited the museum with his parents on the second day of his first trip to Shanghai.
Since opening its doors to the public in late 2001, the museum has grown into one of China's most popular science and education destinations. Through a wide range of science and technology activities and programs, it remains dedicated to boosting public scientific literacy and sparking curiosity in visitors of all ages.
After being closed for over two years for upgrading, the museum reopened early this year. It now also includes two branches: the Shanghai Natural History Museum and the Shanghai Astronomy Museum.
Back in 2001, the proportion of Chinese citizens with scientific literacy was only 1.4 percent, and the figure had risen to 16.74 percent by 2025, according to official data.
As public interest in science has grown, visits to science and technology museums have increasingly become part of everyday life.
For many families, museums have become a popular destination during holidays and weekends, offering children opportunities to expand their knowledge and broaden their horizons.
"The biggest change I feel during my museum visits is that science and technology keep advancing," said Zheng Qina, a post‑90s visitor from Shanghai. She was nine years old when she visited the museum in 2001. Today, her five‑year‑old son is also a regular visitor.
The trend is also extending to international travelers. With China's inbound tourism gaining momentum following eased visa policies, science museums have emerged as an increasingly popular destination for overseas visitors seeking a closer look at the country's technological development.
From January to July this year, the Shanghai Science and Technology Museum and its two branches received nearly 200,000 foreign visitors, accounting for nearly 5 percent of the total visitor numbers.
Francisco and his parents, a family from Italy, were among the visitors on Tuesday. After touring an exhibition on green energy, Francisco said he was deeply impressed by displays showcasing China's nuclear and hydrogen energy technologies.
In recent years, China has made major advances in fields ranging from artificial intelligence and humanoid robots to quantum science, green energy, space technology, life sciences and physical sciences. Science and technology museums offer visitors an opportunity to explore some of these developments through exhibitions and interactive experiences.
For Ni Minjing, director of the Shanghai Science and Technology Museum, a notable change among visitors is that they have shifted from passive recipients of knowledge to more active explorers of science.
"Teenagers in particular now have abundant channels to acquire knowledge. The role of science and technology museums can no longer be merely about resolving doubts," Ni said. "Instead, through diverse forms of science popularization, museums should inspire visitors, especially young people, to think and create."
"We will continue to innovate the way we popularize science, enrich science-related content, promote the spirit of scientists and foster a culture of innovation, so as to spark young people's curiosity, imagination and desire to explore," Ni added.
。
Anthropic又有两张新牌，被提前掀开了！今天，第三方开发者应用和Discord社区中，接连出现了两个陌生的模型ID——claude-mashmallow-eap（棉花糖）claude-melon-eap（甜瓜）
早期实测流出，Marshmallow表现更强，对话自然度反超 Opus 5，Melon则稍逊一筹。不过，它们的综合能力全不是Fable级别，最快下周上线。
B | 就在全网吃瓜的同时，Anthropic内部却先炸出了一颗雷——Fable 5上线两个多月，却遭遇了一场近乎集体的冷落。
Anthropic紧急补牌
Claude新模型曝光这轮更新，似乎比以往来得更急。AI圈大佬们推测，Marshmallow大概率对应Opus级别的迭代升级，可能就是Opus 5.1。而Melon则更接近Sonnet级别。一位网友放出的SVG实测结果，也进一步拉开了两款模型之间的差距。
 |  |
C | 
如果只是正常升级，这件事本身并不奇怪。真正微妙的是，为什么偏偏是现在？毕竟，就在两个多月前，Anthropic才刚刚把自己最强、最贵的一张牌推上牌桌。它就是站在Claude金字塔最顶端的Fable 5。
D |
最强Fable 5，竟然爆冷两个月前，Fable 5顶着Anthropic最强旗舰的光环登场。能力最强，价格最贵，同时还是Claude整个产品线中最顶尖的那个AI。按照行业惯性，新旗舰一发布，大厂往往会迅速迁移，把预算向最新、最强的模型倾斜。这一次，规律在Fable 5身上失效了！
支付平台Ramp追踪了，美国超70000万家企业的Token消费数据，结果发现：Fable 5上线一个月后，只吃掉了企业在Anthropic上6%的Token调用量，支出占比11.4%。如今，两个多月都过去了，FT拿到的最新数据，这一支出份额仍徘徊在11%左右。也就是说，真正愿意把模型切到Fable 5上的开发者、大厂，寥寥无几。对面OpenAI旗舰GPT-5.6 Sol，同期拿走了25%的Token和23%的支出份额。同一个赛场，Anthropic顶配模型份额，只有竞品的四分之一。更尴尬的是，Fable 5的总营收贡献大约只有GPT-5.6 Sol的75%，尽管它每个Token贵了一倍。
贵到什么程度呢？Fable 5的单价是，Anthropic自家Opus 4.8的2倍，是Haiku 4.5的10倍。
E | 虽强，但大多数企业的日常场景，根本用不到那个级别。自家Opus 5，反杀大哥Anthropic另一款模型，才是Fable 5的正面暴击。7月底，Opus 5正式登场，定价只有Fable 5的一半，但在编程、知识工作评测中，反而跑出了更好的成绩。
CursorBench 3.2的测试结果，更直接——Opus 5在最大算力下得分70%，每任务成本$8.23。Fable 5得分70.5%，高了0.5个百分点，但要$17.32。
F | 相当于多花一倍的钱，也仅多跑出0.5%的成绩。企业用脚投票的结果毫不意外，Ramp数据显示，Opus 5上线后，它的支出迅速反超了Fable 5。
但问题在于，如果旗舰模型只是展示柜，那市场凭什么给出2万亿美元的估值？开源从11%到62%，仅四个月不仅如此，开源AI的疯狂扩张，同样让Anthropic倍感压力。Vercel AI Gateway统计显示——4月，开源模型的Token份额只有11%。
G | 到了6月，这个数字暴涨到29%。8月最新数据，直接冲到了62%。短短两个月，从不到三成飙升到超过六成。而它们花掉的钱，不到企业总支出的4%。换句话说，干了大部分活的模型，几乎不花钱。
这就让Fable 5的位置越来越尴尬。
H | 往里看，Opus 5只要一半的价格，就已经能追到它身后；往外看，开源模型还在一轮轮把价格往下打。Fable 5当然可以继续负责冲击能力上限。但Anthropic眼下更缺的，是企业真正愿意长期用、放量用的模型。这么看，Marshmallow（棉花糖）和Melon（甜瓜）突然冒出来，也就没那么意外了。它们要补的，可能正是Claude全家桶里最关键的一块——够强，也得够便宜；能秀肌肉，更得有人买单。
Current article:http://p95.liangkuataosanxuanmiancen.sbs/list_ko5uf1d/h0mho.html
Published on:09:40:09
