Anthropic компанийн дэвшилтэт хиймэл оюун ухааны загвар зах зээлд өрсөлдөх чадвараа алдаж, байгууллагууд илүү хэмнэлттэй шийдлүүдийг сонгох боллоо.
Финтек компани Ramp-ын 70 мянган бизнесийн байгууллагаас цуглуулсан мэдээллээс үзэхэд, Anthropic-ийн орлогод Fable 5 загварын эзлэх хувь ердөө 11 хувьтай байна. Энэ нь хуучин бөгөөд харьцангуй хямд Opus 4.8 загварын хэрэглээтэй харьцуулахад бага үзүүлэлт юм. Fable 5 нь нэг сая оролтын токен тутамд 10 ам.доллар, гаралтын токен тутамд 50 ам.долларын өртөгтэй нь зах зээл дэх хамгийн үнэтэй загваруудын нэгд тооцогдож байна.
Анх долдугаар сарын 01-нд олон нийтэд нээлттэй болсон уг загварыг кибер аюулгүй байдлын өндөр чадамжтай хэмээн сурталчилж байсан ч бодит байдал дээр өндөр үнэ болон хэт мэдрэг хамгаалалтын систем нь хэрэглэгчдийг түлхэх шалтгаан болжээ. Компаниуд нарийн төвөгтэй судалгаа эсвэл хакердах чадвар гэхээсээ илүүтэй өдөр тутмын үйл ажиллагааны хэрэгцээг хангах, зардал багатай хиймэл оюун ухааныг эрхэмлэх болсон нь энэхүү хандлагыг тодорхойлж байна. Нэмж дурдахад, Anthropic компани өөрийн Claude Max 5x болон 20x төлөвлөгөөний токений хязгаарын талаар хэрэглэгчдийг төөрөгдүүлсэн хэмээн зургаадугаар сарын 15-нд шүүхэд гомдол гаргажээ.
Одоогийн байдлаар хиймэл оюун ухааны зах зээлд Moonshot, Z.ai зэрэг Хятадын лабораториудын боловсруулсан хямд, үр ашигтай загварууд хүчтэй өрсөлдөж байна. АНУ болон Хятадын хоорондох технологийн өрсөлдөөн хурцдаж, оюуны өмчийн хулгай, загвар бүтээх үйл явц дахь “distillation” технологийн асуудлаар харилцан буруутгалууд тасрахгүй байна. Хятадын тал өөрийн хүчирхэг загваруудаа гаднын нөлөөнөөс хамгаалах зорилгоор цахим хэрэм босгох боломжийг судалж байгаа нь салбарын нөхцөл байдлыг улам бүр хурцатгаж байна.
Дэлгэрэнгүйг эх сурвалжаас харах
↓Эх сурвалжийг нээх ↓
Fable 5, Anthropic’s poster child turned problem child, is lagging behind competitors as businesses reject the most advanced models in favor of cheaper, more efficient alternatives.
Data published earlier this month by fintech company Ramp shows that total spending on Fable 5 constitutes just 11% of Anthropic’s current revenue from enterprise clients—far less than the older and cheaper Opus 4.8. The data was collected from 70,000 businesses and previously reported by the Financial Times.
NEW from Ramp AI Index: disappointing adoption of Fable 5
We’ve heard several reasons from businesses…mainly Fable 5 is just too expensive.
A model so powerful it was briefly banned, and yet businesses don’t think it’s worth the price. pic.twitter.com/ootF7gn4TK
— Ara Kharazian (@arakharazian) August 12, 2026
Fable 5 was released in June and promoted by Anthropic as a declawed version of Mythos, the much-feared cybersecurity Houdini that had sent a wave of fear reverberating throughout the tech sector. It had only been publicly available for a few days before the Trump administration, citing an obscure export law, ordered that it be blocked to all “foreign persons.” Anthropic was forced to take it offline for all customers as a result. The incident only fueled the belief that the company had built an AI system unlike anything the world had seen up to that point. Fable 5 was re-released to the public on July 01.
But political intrigue and technical prowess don’t have much to do with the plateauing of Fable 5’s adoption, according to the Financial Times. The real reason is much simpler: Businesses are tired of bleeding cash for state-of-the-art models—at $10 per million input tokens and $50 per million output tokens, Fable 5 is one of the priciest models on the market—when less expensive or free models can perform most of the functions they care about just fine. Most companies don’t need a model that can do advanced biomedical research or hack into the NSA mainframe; they just need to be able to handle day-to-day operational tasks.
Complaints about Fable 5 started to be aired online just hours after it was released, with early users saying the model’s hyper-sensitive safeguards made it close to unusable, and that its steep price pointed towards a direction they didn’t want to go in: a future where a small elite can afford the most advanced AI, domineering over a “permanent underclass.” That was followed by aclass action lawsuit, filed June 15, which alleged that Anthropic had been misleading customers about token limits allowed through the company’s Claude Max 5x and 20x plans.
It all points to growing wariness about paying for frontier models with often confusing pricing plans, especially as more affordable alternatives become available. Many of the latter are coming from Chinese labs like Moonshot and Z.ai. (Even the rogue hacking abilities of Moonshot’s AI are more elegant than those of expensive American models.) The “tokenmaxxing” trend has slowly given way to a sober, frugal approach to using AI at work.
The rapid advances among Chinese labs have been chilling conditions in the AI Cold War between the US and China. Anthropic, OpenAI, and the Trump administration have accused Chinese AI labs of essentially stealing their IP to train their models via a legally ambiguous process of international, large-scale distillation. And China has reportedly weighed the possibility of building a firewall around its most powerful models, cutting off all foreign access.
