[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-anthropic-fable-5-plateau-11-percent":3,"news-related-e1724d68-bf0d-4b3f-8047-147796d5d52e":38},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"e1724d68-bf0d-4b3f-8047-147796d5d52e","Ramp 8 月指数:Fable 5 企业份额停滞 11%,OpenAI 旗舰跑赢两倍","Ramp 追踪 7 万家美国企业：Anthropic 旗舰 Fable 5 仅占 11.4% 美元份额、6% token 份额；同期 OpenAI GPT-5.6 Sol 占 23% 与 25%。Fable 5 每百万 token 约 10 美元，是 GPT-5.6 两倍，Opus 5 已反超。","两个月前，Anthropic 把号称最强的 Fable 5 推向企业市场，定价每百万 token 约 10 美元，是当时 OpenAI GPT-5.6 Sol 的两倍左右。Ramp 在其 8 月指数里给出了一组相当难看的数字：基于 7 万家美国企业的实际支付数据，Fable 5 在 Anthropic 内部仅占 11.4% 的美元份额和 6% 的 token 份额，且增长已趋于停滞；作为对比，GPT-5.6 Sol 在 OpenAI 内部拿下了 23% 的美元份额与 25% 的 token 份额。\n\n更刺眼的数字是收入侧。Fable 5 在 7 月只跑出了 GPT-5.6 Sol 大约 75% 的营收。这意味着：愿意为最强模型付顶价的客户群体明显比一年前窄了。Ramp 经济学家 Ara Kharazian 把矛头直接指向价格和数据留存要求：\"GPT-5.6 Sol 真的很强，越来越多开发者把它当默认；Fable 5 在采用率和真实业务落地上都让人失望，价格和 30 天数据留存限制都是拖累。\"\n\n事情还没完。7 月底 Anthropic 又发了一款定价更低的 Claude Opus 5，不到两个月，Opus 5 在企业支出上就已经反超自家旗舰 Fable 5。客户没有涌向更贵的最强模型，反而被更便宜的那一款拉走。这彻底打破了企业默认选最强模型的惯性。\n\nAccel 合伙人 Miles Clements 对 FT 的评论一针见血：客户第一反应就上手旗舰模型的年代\"不是可持续的时代\"。\"大多数场景根本不需要在前沿跑。\"换句话说，前沿 AI 实验室花几十亿美元把模型堆得越来越复杂，但企业支付端对\"前沿溢价\"的容忍度被 Fable 5 钉死了新上限——大约每百万 token 10 美元，再贵就不买了。\n\n但这并不意味着 Anthropic 输了。在 Ramp 的更宽口径上，Anthropic 仍以 43.5% 的美国企业渗透率领先 OpenAI 的 39.7%，7 月年化营收 Run Rate 已经冲到 650 亿美元，5 月时还是 470 亿。问题不在用户基数，而在客单价结构：旗舰没扛起重任，增长靠的是中端 Opus 和低价模型。\n\n这件事最值得思考的一点，是它正在改变前沿 AI 实验室的商业模型剧本。过去十年，故事是\"模型越大越强，订阅费按 token 单价线性上涨\"；Ramp 的数据第一次给出反向信号——当客户已经能用 70% 价格的模型完成 90% 的任务时，前沿溢价就不再是免费选项。下一步，如果 OpenAI、xAI、Google DeepMind 跟进做出\"低价旗舰\"对冲，Fable 5 这种定价策略会变成昂贵的试错。\"最强模型\"不再是商业故事的终点，而是要回答\"谁愿意为它付钱、付多少\"。","https:\u002F\u002Fwww.ft.com\u002Fcontent\u002F5ee49718-c258-4f01-aa32-7e5b76ae5245","8fc22e42-e2dd-442b-84e4-77eb68061c39",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"23544f6a-eea1-4f05-aa8d-749ca862d5d2","anthropic",{"id":19,"name":20,"slug":20,"description":14,"color":14},"dca4d0ab-7994-43a7-839e-7756fc77344a","claude",{"id":22,"name":23,"slug":23,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"8f65a905-4ac1-430d-b432-82190003bba9","en","Ramp August Index: Fable 5 stalls at 11% of Anthropic spend, OpenAI flagship pulls double","Ramp tracked 70,000 US businesses: Anthropic's flagship Fable 5 took just 11.4% of dollar share and 6% of token share, while OpenAI's GPT-5.6 Sol captured 23% of dollars and 25% of tokens. Fable 5 prices around $10 per million tokens, roughly 2x GPT-5.6 Sol. The cheaper Claude Opus 5 has already overtaken Fable 5 in enterprise spend.","Two months after Anthropic pushed its most powerful model, Fable 5, into the enterprise market at roughly $10 per million tokens—about double OpenAI's GPT-5.6 Sol—Ramp's August index delivered a brutal number: across 70,000 US businesses, Fable 5 captured only 11.4% of Anthropic dollar spend and 6% of token spend, with growth flat-lining. By comparison, GPT-5.6 Sol took 23% of OpenAI dollar share and 25% of token share.\n\nThe revenue picture stings more. In July, Fable 5 generated only about 75% as much revenue as GPT-5.6 Sol. The pool of customers willing to pay top dollar for the strongest model has visibly narrowed in the past year. Ramp economist Ara Kharazian pointed the finger directly at price and data-retention terms: \"GPT-5.6 Sol is really good, increasingly the default for developers. Fable 5, meanwhile, disappointed both in adoption and real-world application given price + 30-day data-retention requirements.\"\n\nIt does not stop there. At the end of July, Anthropic shipped Claude Opus 5 at a lower price. Within two months, Opus 5 overtook its own flagship Fable 5 in enterprise spend. Instead of flocking to the most expensive top-tier model, customers drifted toward the cheaper option. The old default—enterprise buyers reaching first for the strongest model—broke in public.\n\nAccel partner Miles Clements summed it up for the FT: the era when customers reached first for flagship models \"was not a sustainable era.\" \"Most people don't need to operate at the frontier.\" In other words, frontier AI labs are still sinking tens of billions of dollars into ever larger and more complex models, but the enterprise buyer's tolerance for a frontier premium just got capped by Fable 5 at roughly $10 per million tokens. Above that, the purchase evaporates.\n\nThat said, Anthropic is not losing. On Ramp's broader adoption measure, Anthropic still leads OpenAI at 43.5% of US tracked businesses versus 39.7% in July, and its annualised revenue run rate hit $65 billion that month, up from $47 billion in May. The issue is not user base but unit-economics structure: the flagship did not carry the load, and growth now rides on mid-tier Opus and the lower-priced models.\n\nThe deeper takeaway is that Ramp's data quietly rewrites the frontier-lab business-model playbook. For the past decade the story ran: bigger model, stronger model, linearly rising per-token price. Fable 5 is the first hard counter-signal—when customers can hit 90% of their tasks with a 70%-priced model, the frontier premium stops being free. If OpenAI, xAI and Google DeepMind follow up with their own \"cheap flagship\" hedge, the Fable 5 pricing strategy will turn into an expensive experiment. \"The strongest model\" is no longer the endpoint of the business story; the real question becomes \"who is willing to pay for it, and how much.\"","anthropic-fable-5-plateau-11-percent","2026-08-25T06:00:00Z","2026-08-25T03:03:58.495689Z","2026-08-25T03:03:58.495698Z",true,"agent",45,{"items":39},[40,45,50,55,60,65],{"id":41,"title":42,"news_slug":43,"published_at":44},"39724847-fdc9-4199-ac46-311e7b49d385","Ramp 数据复盘 Fable 5:旗舰上市两月仅占企业 Anthropic 支出 11%,70 倍价差压住前沿模型溢价","ramp-data-fable-5-adoption-plateaus","2026-08-26T08:00:00+00:00",{"id":46,"title":47,"news_slug":48,"published_at":49},"1051d676-8ed9-4448-b0d5-8db4b844f41f","Claude Fable 5 上线两个月,为什么企业只把 11% 的账单花给最强模型","claude-fable-5-11-percent-anthropic-spend","2026-08-25T06:00:00+00:00",{"id":51,"title":52,"news_slug":53,"published_at":54},"454f9530-20d7-428c-82d9-9175fa5b883a","Claude 推黎曼 zeta 下界到 67.2%：60 subagent + Lean","claude-zeta-bound-67-percent-multi-agent-lean","2026-08-17T07:00:00+00:00",{"id":56,"title":57,"news_slug":58,"published_at":59},"db4ffdac-3734-41a2-8e32-67feaa7341bd","Claude 冲击黎曼猜想\"失败\",却顺手改写了 37 年没人动过的数学纪录","claude-riemann-zeta-67-percent-record","2026-08-16T23:30:00+00:00",{"id":61,"title":62,"news_slug":63,"published_at":64},"a7146291-e849-42c3-acbd-2b50627d5332","Claude Opus 4.8 发布：41天极速迭代，Dynamic Workflows 重塑Agent协作范式","claude-opus-4-8-41-day-dynamic-workflows","2026-05-29T04:00:00+00:00",{"id":66,"title":67,"news_slug":68,"published_at":69},"97c97b9c-e6e4-4982-aa57-0c0da814fb19","Anthropic 的欧盟答卷四小时即被撕开：Claude 文本水印为什么怕改写","claude-synthid-70-percent-threshold-bypass","2026-08-21T08:00:00+00:00"]