Anthropic Has the Most Powerful Model on the Market. So Why Don’t Companies Want It?

Anthropic Has the Most Powerful Model on the Market. So Why Don’t Companies Want It?

Ondřej Barták
Ondřej Barták
Entrepreneur and Programmer
26. 8. 2026
3 minutes reading · 1 views
Listen to the article
Audio version of the article
Anthropic Has the Most Powerful Model on the Market. So Why Don’t Companies Want It?

American companies have access to the most powerful model Anthropic has ever built, yet they are largely leaving it unused. Two months after its launch, Fable 5 accounts for only about eleven percent of everything companies spend with Anthropic. This is according to data from payment company Ramp, which processed spending by seventy thousand American businesses.

Until now, one rule held true. When a technology company unveiled a new flagship offering, corporate customers switched to it almost automatically. With the Claude model, however, this pattern has broken down for the first time, with its share of spending stalling at eleven percent. According to Ramp, that represents just over eleven percent of dollars spent and approximately six percent of tokens processed. Companies reserve the model for exceptional tasks and send routine work elsewhere. 

Cheaper models are taking over the work

The Fable 5 model entered the market in June and immediately became entangled in politics. The Donald Trump administration forced Anthropic to withdraw the model, citing national security, and it returned on July 1 after receiving government approval. Today, however, politics plays a relatively minor role in model selection. Anthropic's analysts and investors agree that the main factors are Fable 5's high price and the fact that older models can handle most business requirements.

This is most evident in where demand has shifted. The Opus 4.8 model, a less powerful and more affordable option, jumped from fifteen percent at the beginning of June to more than half by the end of July. The Opus 5 model, which Anthropic released at the end of July, then overtook the flagship Fable 5 in corporate spending shortly after its launch. Customers therefore did not leave for competitors; they simply chose a cheaper option from Anthropic's own lineup.

At the same time, freely available models from China are moving into the same space, further intensifying price pressure. Companies can therefore use either older Anthropic models or open-source Chinese competitors as cheaper alternatives.

Price now matters more than performance

The difference in cost is substantial. The Fable 5 model costs around ten dollars per million tokens, approximately twice the price of OpenAI's GPT-5.6 Sol model, and in July it generated only three-quarters of the revenue earned by its competitor's flagship model. After a lackluster start to the year, OpenAI has regained momentum, with its annualized revenue rising by thirty-five percent during the current quarter to exceed forty billion dollars. This recovery was driven by the July launch of GPT 5.6, which is significantly cheaper than Fable 5.

The period when customers automatically reached for the most powerful model could not last long. Most people simply do not need to operate at the limits of what is possible. 

Anthrropic is still doing well

Weaker interest in the Fable 5 model does not mean the company is no longer doing well. In July, Anthropic led in the number of paying American companies tracked by Ramp, with forty-three percent compared with just under forty percent for OpenAI. Its annualized revenue also reached sixty-five billion dollars, up from forty-seven billion in May. Compared with revenue of nine billion dollars at the end of 2025, this is more than a sevenfold increase. 

According to the Financial Times, six Anthropic investors expect the company to go public in October at a valuation of at least two trillion dollars. Their models predict that annualized revenue will reach between one hundred and one hundred twenty billion dollars by the end of the year. These are, however, investor expectations, not figures endorsed by the company itself. Its internal forecast targets two hundred billion dollars in revenue only in 2028.

Category:AI
Did you enjoy this article?
Discover more interesting posts on our blog
Back to blog

Related posts

Mistral enters the Saudi market with its open modelsMistral enters the Saudi market with its open models
Mistral has signed a deal worth hundreds of millions of euros with Saudi Arabia's HUMAIN. Together, they aim to develop Arabic-language models and advance sovereign AI in sensitive sectors.
4 min read
26. 8. 2026
What If an AI Model Rebels and Escapes? Leading Labs Like Meta Have Almost No SafeguardsWhat If an AI Model Rebels and Escapes? Leading Labs Like Meta Have Almost No Safeguards
A new assessment of five leading AI labs reveals weaknesses in their controls. None has a public plan for stopping a model that breaks free from human oversight.
6 min read
26. 8. 2026
World Models vs. LLMs: How They Differ and Why They’re Key to the Future of Physical AIWorld Models vs. LLMs: How They Differ and Why They’re Key to the Future of Physical AI
LLMs understand words, but world models can predict the consequences of actions in the physical world. This difference could shape the future of robotics.
7 min read
26. 8. 2026
Přihlaste se k odběru našeho newsletteru
Zůstaňte informováni o nejnovějších příspěvcích, exkluzivních nabídkách, a aktualizacích.
CodedTrip

Operated by CodedTrip LLC, USA.

YouTube
TikTok