Key Summary
- Ramp data: Astra at ~13%, Fable at 8%
- How one new model shifted the spending landscape
- Rework required when switching models
- Counting the tasks that would stop if you switched providers
According to data compiled by enterprise expense management platform Ramp, OpenAI's 'GPT-6 Astra' now holds about 13% of enterprise AI spend, overtaking the 'Claude Fable' line at 8%. On the model routing platform OpenRouter as well, spending on OpenAI's product lineup surpassed Anthropic's for the first time in two and a half years.
What has changed in the enterprise AI market
- September 3, 2026: OpenAI released 'GPT-6 Astra,' featuring significantly enhanced computer control and security capabilities.
- September 12: Anthropic CEO Dario Amodei argued that the pace of frontier model development should be slowed.
- September 19: Reuters reported, citing three sources, that Anthropic is carefully weighing whether to release a new AI model before its planned IPO.
As of the report, Anthropic's next-generation model remains in the comprehensive safety evaluation stage. This comes just a week after CEO Amodei raised the case for slowing down development, while OpenAI CEO Sam Altman—who agreed with the call for a measured pace—has signaled a 'major new product' launch next week.
Implications for Korean Enterprise AX
These figures reflect spending data compiled independently by Ramp and OpenRouter. For Korean companies looking to use this as a basis for model selection, the sensible approach is not to read it as a market ranking, but to conduct a separate comparison of two or more models under identical conditions using their own workflows and Korean-language documents.
This case illustrates how a single new model can reshape the landscape of enterprise AI spending. Anthropic is reviewing whether to release a new model, while OpenAI has signaled an upcoming product launch. If a company's operational systems are built around a single model, every model switch requires starting prompt design, evaluation, and security review from scratch.
What to check first to prepare for a model switch
- Compile a list of which models your current AI workflows are connected to.
- Document, for each workflow, the rework required when switching models—prompt revisions, output evaluation, and security review.
- Build an evaluation set using samples from your own workflows to compare multiple models under the same criteria.
There is one thing worth checking this week: count how many of your active AI workflows would stop functioning if you switched providers.