Anthropic’s Claude Fable 5 was launched as one of the most capable AI models available for demanding coding, research and enterprise work, but early business spending data suggests that companies may be more selective about paying premium prices for frontier AI than the industry’s biggest headlines imply. Ramp’s July data shows that spending on Fable 5 reached only about 75% of the level generated by OpenAI’s GPT-5.6 Sol, despite Anthropic maintaining a larger overall business customer base.
The figures point to an emerging divide in the enterprise AI market. Companies are clearly increasing their spending on artificial intelligence, particularly for coding agents and other advanced workflows, but that does not necessarily mean they are willing to pay any price for the most capable model. Fable 5 costs significantly more than several competing models, while GPT-5.6 Sol offers broadly comparable frontier performance at lower prices. The result suggests that the next phase of enterprise AI adoption could be driven as much by cost efficiency as by raw model intelligence.
Fable 5’s Enterprise Adoption Falls Short of OpenAI
Anthropic launched Claude Fable 5 in June as its most advanced generally available model at the time.
The model was designed for long-running coding projects, complex knowledge work, financial analysis, research and autonomous AI-agent workflows.
However, Ramp’s July data shows that Fable 5 did not generate as much business spending as OpenAI’s GPT-5.6 Sol.
Fable 5 generated approximately 75% of the spending associated with GPT-5.6 Sol among the businesses tracked by Ramp.
| Enterprise AI Spending Indicator | July 2026 |
|---|---|
| Fable 5 spending vs GPT-5.6 Sol | ~75% |
| Businesses using Anthropic | 43.5% |
| Businesses using OpenAI | 39.7% |
| Top 1% AI spend per employee | ~$7,400/month |
| Top 10% AI spend per employee | ~$611/month |
| Median AI spend per employee | ~$11/month |
Ramp’s dataset covers more than 70,000 small and midsize U.S. businesses, making it a useful indicator of how companies are actually deploying and paying for AI rather than simply what executives say they plan to spend.
Anthropic Still Has More Business Customers
The Fable 5 spending figures are notable because Anthropic continues to have a higher overall business penetration rate than OpenAI in Ramp’s data.
Approximately 43.5% of businesses tracked by Ramp were using Anthropic products, compared with 39.7% for OpenAI.
That means Anthropic has a larger customer footprint, but OpenAI’s flagship model was generating more spending.
Customer Reach vs Premium Spending
Anthropic
43.5% of businesses
↓
Broader business adoption
↓
Fable 5
~75% of GPT-5.6 Sol spending
OpenAI
39.7% of businesses
↓
Smaller overall business reach
↓
GPT-5.6 Sol
Higher premium-model spending
The difference suggests that customer count and high-end AI spending are becoming two different measures of enterprise success.
Why Fable 5 May Be Struggling to Capture Premium Spending
Price is one of the clearest explanations.
Fable 5 is priced at $10 per million input tokens and $50 per million output tokens.
GPT-5.6 Sol is priced at $5 per million input tokens and $30 per million output tokens.
That makes Fable 5 twice as expensive for input tokens and significantly more expensive for output tokens.
| Frontier Model | Input / 1M Tokens | Output / 1M Tokens |
|---|---|---|
| Claude Fable 5 | $10 | $50 |
| GPT-5.6 Sol | $5 | $30 |
| Fable 5 premium vs Sol | 2X | ~1.67X |
For businesses running millions or billions of tokens, these differences can quickly become substantial.
The more an AI system is used, the more important the cost of every additional token becomes.
Frontier AI Is Becoming a Cost Calculation
The early AI market was dominated by a simple question:
Which model is smartest?
Enterprise customers are increasingly asking a different question:
Which model delivers the best result for the lowest total cost?
This shift matters because businesses rarely use AI for a single question.
An enterprise coding agent can make hundreds of model calls during one task.
A research agent can process thousands of documents.
A customer-support system can generate millions of responses.
Even a small difference in per-token pricing can therefore produce a large difference in the final bill.
Enterprise AI Decision
Model capability
+
Reliability
+
Speed
+
Token consumption
+
Price
↓
Cost per completed task
That final figure may matter more to companies than the model’s position on a benchmark leaderboard.
GPT-5.6 Sol Changed the Economics
OpenAI’s GPT-5.6 Sol arrived roughly one month after Fable 5.
The timing was significant because OpenAI offered a frontier model at a lower price.
Independent Artificial Analysis testing also found GPT-5.6 Sol close to Fable 5 on its Intelligence Index while costing roughly one-third as much per task under the tested configurations.
This created a difficult commercial environment for Anthropic.
If two models deliver similar results on a particular business workflow, companies have a strong financial incentive to choose the cheaper option.
The New Frontier AI Equation
Fable 5
Higher capability
+
Higher price
↓
Premium positioning
GPT-5.6 Sol
Comparable frontier capability
+
Lower price
↓
Better price-performance
The competition is therefore moving beyond intelligence alone.
Businesses Are Spending More on AI, But Not Equally
Ramp’s data does not show that companies have stopped spending on AI.
Quite the opposite.
AI is reportedly the fastest-growing category of technology spending in Ramp’s dataset.
The important point is that spending is highly concentrated.
The top 1% of AI-spending businesses were spending a median of roughly $7,400 per employee per month in July.
The top 10% were spending approximately $611 per employee.
The median business was spending only around $11 per employee.
| Business Group | Approx. AI Spend per Employee |
|---|---|
| Top 1% | ~$7,400/month |
| Top 10% | ~$611/month |
| Median business | ~$11/month |
The enormous gap indicates that enterprise AI adoption is still highly uneven.
A relatively small group of companies is spending aggressively, while the typical business remains cautious.
The AI Spending Curve Is Extremely Uneven
The numbers reveal a sharp divide between AI leaders and the broader business market.
For the top 1% of companies, AI spending can represent a major technology investment.
For the median company, AI spending remains extremely small.
Enterprise AI Adoption
AI leaders
Thousands of dollars per employee
↓
Large-scale AI deployment
↓
Agents and automation
Typical businesses
Low AI spending
↓
Experimentation
↓
Selective adoption
This suggests that the market is still in an early adoption phase.
The headline numbers from the biggest AI buyers should therefore not be interpreted as evidence that every company is rapidly deploying expensive frontier models.
Coding Agents Are Driving High-End Spending
One of the biggest drivers of AI spending is software development.
Coding agents can perform tasks that previously required significant amounts of developer time.
They can inspect repositories, write code, run tests, identify bugs and make changes.
Because these systems can make large numbers of model calls, they also consume substantial amounts of AI inference.
That makes coding one of the areas where model pricing can have a direct impact on business economics.
Coding Agent Economics
Developer gives task
↓
AI agent analyzes codebase
↓
Generates code
↓
Runs tests
↓
Finds errors
↓
Makes corrections
↓
Runs tests again
↓
Completes task
Each additional reasoning and tool-use cycle can increase token consumption.
A more expensive model therefore needs to provide significantly better results to justify its additional cost.
Fable 5 Was Built for Long-Running Work
Anthropic positioned Fable 5 specifically for ambitious, long-running tasks.
The company says the model can work for days in an agent harness such as Claude Code, plan across multiple stages, delegate to sub-agents and check its own work.
This makes the pricing issue even more important.
A model used occasionally for difficult questions can justify a premium price relatively easily.
A model operating continuously as an enterprise agent can generate a much larger bill.
Long-Running Agent
Initial task
↓
Planning
↓
Sub-agent calls
↓
Tool use
↓
Reasoning
↓
Verification
↓
Additional corrections
↓
Final output
The longer the workflow, the more sensitive businesses become to token economics.
Companies May Not Need the Most Powerful Model for Every Task
Another factor limiting premium-model adoption is model specialization.
Companies do not necessarily need their most powerful AI model for every task.
A business might use a frontier model for difficult research while using a cheaper model for customer support, document summarization or routine coding.
This creates a model-routing strategy.
Enterprise Model Routing
Complex research
↓
Frontier model
Advanced coding
↓
Premium coding model
Routine coding
↓
Lower-cost model
Summarization
↓
Fast inexpensive model
Simple customer questions
↓
Small model
This approach can reduce total AI spending while preserving access to expensive models when they actually provide additional value.
Fable 5’s Pricing Creates a Higher Bar
At $10 per million input tokens and $50 per million output tokens, Fable 5 sits at the expensive end of the mainstream frontier-model market.
That does not necessarily make it unattractive.
For certain tasks, the model’s ability to reason through complex problems, work autonomously and verify its own output could justify the premium.
But businesses need to measure that benefit.
If Fable 5 saves an engineer several hours on a difficult task, the additional model cost may be insignificant compared with the value created.
If it produces only marginally better results than a cheaper model, the economics change.
Price-Performance Is Becoming the New AI Battleground
The latest generation of AI models is making raw benchmark leadership less decisive.
A model that is slightly better but significantly more expensive may lose to a model that is nearly as capable but much cheaper.
This is particularly true for companies deploying AI at scale.
AI Model Competition
Generation 1
Who has the smartest model?
↓
Generation 2
Who has the best reasoning?
↓
Generation 3
Who has the best agents?
↓
Current competition
Who delivers the best intelligence per dollar?
The industry is increasingly moving toward the fourth question.
OpenAI’s Lower-Cost Strategy Could Pressure Anthropic
OpenAI’s pricing for GPT-5.6 Sol creates additional pressure on Anthropic.
If OpenAI can deliver comparable performance at a lower cost, enterprise customers have an incentive to test or migrate workloads.
This does not mean Anthropic is losing its enterprise advantage.
Its larger customer penetration indicates strong demand for Claude products.
But the spending data suggests OpenAI is becoming more competitive at the high end of enterprise AI.
Anthropic Has Other Advantages
Fable 5’s spending performance should not be interpreted as evidence that Anthropic’s enterprise business is weak.
Anthropic continues to have a broad customer base and a strong reputation for coding and professional workflows.
Its models are available through major cloud platforms, including Amazon Web Services, Google Cloud and Microsoft infrastructure.
The company also has deep relationships with software developers and enterprises.
Fable 5 is therefore only one component of Anthropic’s overall commercial strategy.
The Fable 5 Launch Was Also Unusually Disrupted
The model’s adoption trajectory was affected by a difficult launch.
Anthropic temporarily suspended access to Fable 5 and its more powerful Mythos 5 after a U.S. government export-control directive.
Access was restored in July after the restrictions were lifted.
Anthropic then changed how Fable 5 was offered to subscribers, moving toward usage-based credits because of capacity constraints.
These events may have affected early adoption and spending.
Fable 5 Launch Timeline
June 9
Fable 5 launches
↓
June 12
Access suspended following U.S. government directive
↓
June 30
Export controls lifted
↓
July 1
Global access restored
↓
July
Usage-based access and premium pricing
↓
July spending data
OpenAI’s GPT-5.6 Sol generates higher premium-model spending
The unusual launch conditions make it difficult to conclude that pricing alone explains the spending gap.
Capacity Constraints Also Matter
Anthropic initially warned that demand for Fable 5 would be difficult to predict.
The company rolled out subscription access conservatively and later moved toward usage credits.
This indicates that supply and infrastructure constraints were important factors in the model’s availability.
A company cannot generate unlimited revenue from a model if customers cannot access it reliably.
That makes infrastructure capacity another important part of the enterprise AI equation.
Is Corporate Willingness to Pay Actually Hitting a Ceiling?
The current evidence suggests a more nuanced answer.
Companies are clearly willing to spend large amounts on AI.
The top AI-spending businesses are spending thousands of dollars per employee each month.
But there appears to be a ceiling on what companies are willing to pay for incremental model capability.
If a $50-per-million-token model is only slightly better than a $30-per-million-token model for a particular workflow, businesses may choose the cheaper option.
That could make the market increasingly sensitive to price-performance ratios.
The Enterprise AI Market May Split Into Tiers
The future enterprise AI market could develop into several distinct spending tiers.
| Tier | Typical Use | Model Strategy |
|---|---|---|
| Basic | Summaries, simple chat | Low-cost models |
| Productivity | Office and coding assistance | Mid-range models |
| Advanced | Complex coding and analysis | Frontier models |
| Critical | Research, finance, specialized work | Premium frontier models |
| Autonomous | Long-running agents | Best model based on ROI |
This structure would allow companies to spend heavily where AI produces measurable value while keeping routine workloads inexpensive.
What This Means for AI Model Companies
AI labs may increasingly need to prove economic value rather than simply technical superiority.
A model launch will need to answer several questions:
- How much better is it?
- How much does each task cost?
- How many tokens does it require?
- How fast does it respond?
- Can it complete tasks with fewer retries?
- How much human labor does it replace or accelerate?
- Can businesses deploy it reliably at scale?
The strongest model may not necessarily become the most commercially successful model.
Key Numbers at a Glance
75%
Approximate Fable 5 spending compared with GPT-5.6 Sol in Ramp’s July data
43.5%
Share of tracked businesses using Anthropic products
39.7%
Share using OpenAI products
$10
Fable 5 input price per million tokens
$50
Fable 5 output price per million tokens
$5
GPT-5.6 Sol input price per million tokens
$30
GPT-5.6 Sol output price per million tokens
~$7,400
Median monthly AI spending per employee among the top 1% of AI-spending businesses
~$611
AI spending per employee among the top 10%
~$11
Median AI spending per employee
70,000+
Small and midsize U.S. businesses represented in Ramp’s dataset
What Investors Will Watch
The most important issue for Anthropic will be whether its broader enterprise adoption translates into higher spending on its most advanced models.
Investors will also watch whether customers shift workloads between Fable 5, lower-cost Claude models and competing products.
For OpenAI, the key question will be whether GPT-5.6 Sol’s lower price can convert its growing premium-model spending into broader enterprise adoption.
The competition will increasingly be measured through actual customer spending rather than benchmark scores alone.
What This Means for Corporate AI Budgets
Companies appear to be entering a more disciplined phase of AI spending.
The initial period of experimentation is giving way to questions about return on investment.
Executives increasingly need to know whether an AI system saves employees time, increases revenue, reduces costs or enables work that was previously impossible.
This favors models that combine strong performance with predictable costs.
Looking Ahead
Fable 5’s slower premium-model spending compared with GPT-5.6 Sol does not mean businesses are losing interest in frontier AI. Instead, the data suggests that companies are becoming more selective about where they deploy the most expensive models. Anthropic still has a larger share of businesses using its products in Ramp’s dataset, but GPT-5.6 Sol generated higher spending among the companies tracked. The combination of Fable 5’s $10/$50 per-million-token pricing and increasingly capable lower-cost alternatives is forcing enterprises to consider whether the additional intelligence is worth the additional expense.
The broader implication is that the enterprise AI market may be approaching a price-performance inflection point. Businesses appear willing to spend heavily when AI can deliver measurable productivity gains, particularly through coding agents and autonomous workflows, but they may not automatically choose the most expensive frontier model. As models become more similar in capability, the winners could be determined by cost per completed task, reliability, speed and return on investment rather than benchmark leadership alone.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
