Interest in Fable 5 stalls as companies realize they don't need expensive models

Corporate budgets put a leash on expensive models

A new report from payments platform Ramp highlights an unexpected trend. Despite Anthropic releasing its best and most expensive model, Fable 5, companies refuse to spend on it across the board. Expenditure on Fable 5 accounts for only 11 % of the total amount customers spend on Anthropic’s tools, and that share has stopped growing.

It’s not that the model doesn’t work. Customers have simply learned to be smart about routing. They call the flagship model only for the most complex tasks and push routine agendas (categorization, basic extraction, translation) to cheaper alternatives. Often, these are Anthropic’s own older models or open-weight models from Chinese competitors.

The breakdown of a megalomaniac business model

If this enterprise behavior becomes the standard, it poses a major threat to the "go-big-or-go-home" strategy. Leading AI labs (like Anthropic and OpenAI) are pouring billions of dollars into training massive models under the assumption they can charge a premium for them. However, according to Miles Clements of the VC fund Accel (which invested $1 billion in Anthropic), most companies simply do not need to operate at the absolute frontier of capabilities.

Fable 5 also had a rough start. Its launch in June was mired in a security controversy over its purported ability to launch cyberattacks (a narrative Anthropic itself helped fuel), leading to a temporary suspension of access for foreign customers by the US administration. While those restrictions were eventually lifted, the expected surge in adoption never materialized.

OpenAI capitalizes on volume

While spending on Anthropic's flagship stagnates and its annualized revenue run-rate (ARR) lingers at $65 billion (falling $15 billion short of investor expectations), OpenAI is attacking from the bottom. According to the FT, the ChatGPT maker has pushed its ARR to $40 billion, driven primarily by the new GPT-5.6 model, which is significantly cheaper to use than Fable 5.

However, Alex Imas from Google DeepMind points out that flat spending on one specific model isn't necessarily a catastrophe. If cheaper models act as a complement and total spending across all models grows, the overall value of the ecosystem to the customer increases. Anthropic has essentially started profiting from what companies consider "good enough."

The future depends on margin control

Forecasts remain cautious. Ramp economist Ara Kharazian notes that predicting trends in this sector is virtually impossible. Previous indicators pointed to Anthropic’s dominance, but the unexpected quality of OpenAI's new, cheaper model quickly flipped the market dynamics.

Lilith's verdict

Customers have stopped buying a Rolls-Royce for trips to the grocery store. For AI labs, this means they can no longer just sell raw power; they have to start selling actual efficiency.