Foyer, an American AI startup, stated that instead of purchasing enterprise-level coding services from OpenAI and Anthropic, it allows employees to use individual subscription accounts. This approach keeps its monthly AI coding tool expenses at around $3,000, significantly lower than the tens of thousands of dollars that could be incurred under an enterprise API billing model.
Personal packages cover high-frequency usage
Foyer primarily develops an AI browser tool and an AI companion app. CEO Pratyush Rai told Business Insider that AI is widely used internally; finance and marketing staff use the tools to write internal applications, while developers continuously test new versions of Anthropic's Claude Code and OpenAI's Codex.
Foyer's CTO, Siddhartha Saxena, said that if he used the pay-as-you-go OpenAI enterprise account, his Codex usage fees for April alone could have reached $4,000; however, the $200 personal plan covered all his needs for the month. Rai also stated that Foyer's approximately 25 employees' combined monthly expenses for OpenAI and Anthropic personal accounts were about $3,000, while switching to the enterprise API solution could have raised their monthly bills to $30,000 to $40,000.
- Codex individual plan costs approximately $200.
- Total team individual account expenditures are approximately $3,000 per month.
- The estimated cost of an enterprise API is approximately $30,000 to $40,000 per month.
AI tools reduce development manpower
Foyer previously relied on about 20 people to maintain its browser extension product, Merlin AI, and grew it to 900,000 Chrome users. Rai stated that the product line can now be maintained with just 3 developers working on AI coding tools.
The saved manpower was invested in the new product, Thine. This app continuously records audio around the user via smartphone, attempting to provide the AI system with a more complete contextual memory. These functions require speech-to-text processing and the management of large amounts of user data, which Foyer says also heavily rely on AI tools.
Rai said that two or three years ago, this type of work might have required 50 people to complete; now the company can get it done with about 15 developers.
The enterprise version still has security and management advantages.
The report noted that individual plans are cheaper, but enterprise plans typically offer more management capabilities, including stronger data isolation, team access control, and usage visibility. An Anthropic spokesperson stated that individual plans are suitable for small teams, but enterprise customers usually choose enterprise plans due to their security, governance, and visibility needs. OpenAI did not respond to requests for comment.
Foyer management also stated that another benefit of individual subscriptions is the flexibility to switch. With the new model launched, teams can more quickly adapt to different service providers and tiers without being tied to a single corporate purchase plan.
Pay attention to the continued decline in token costs.
Foyer is closely monitoring token price fluctuations. Rai believes that as semiconductor performance improves and more open-source models are released, the unit token cost of large models may continue to decline. If this trend continues, even if OpenAI and Anthropic tighten subsidies for individual plans in the future, the total cost for enterprises on a pay-as-you-go basis may also decrease significantly.
He predicts that current pay-as-you-go bills of $30,000 to $40,000 could potentially drop to the $2,000 to $3,000 range in the future. This would also create space for AI applications like Thine that have higher computational demands.
Saxena said that with improved coding tools and lower current costs, the team has consumed far more tokens this year than in the past.
