Наталя ХандусенкоAI Eng
9 July 2026, 16:45
2026-07-09
Startup Accidentally Spent $30,000 on AI Tokens in a Month
In 2024, Sartak Dhawan and his friend founded Turbo AI, an AI learning tool. They dropped out of college last year, and it seems like it was a good decision, as the company has already crossed the $13 million mark in revenue.
There was one month this year when the startup accidentally spent about $30,000 on Claude Code tokens, and Sartak said it was embarrassing. However, he doesn’t consider that month a complete mistake because he learned from it.
In 2024, Sartak Dhawan and his friend founded Turbo AI, an AI learning tool. They dropped out of college last year, and it seems like it was a good decision, as the company has already crossed the $13 million mark in revenue.
There was one month this year when the startup accidentally spent about $30,000 on Claude Code tokens, and Sartak said it was embarrassing. However, he doesn’t consider that month a complete mistake because he learned from it.
When Turbo AI was launched, the AI models for generating code were pretty terrible, so Sartak wrote the code himself from scratch. As the tools developed, the startup’s spending on AI for programming also grew. Now, a team of 10 people spends an average of about $20,000 per month.
"We are calm about spending on AI as long as it delivers results," Sartak said in an interview with Business Insider.
But in April, Turbo AI's bill for AI tokens jumped to around $30,000. It was a month of active feature releases, and the costs partly reflected this.
The second reason for the increased costs: Sartak worked at Claude in a fast-paced mode without even realizing it.
“This setting in Claude Code makes work visually faster, but also costs significantly more per token. When I turned on fast mode and left it active, my score skyrocketed,” added the Turbo AI co-founder.
Now the team turns on fast mode only when they are working in pairs and signal latency is critical; otherwise it turns it off.
“Exiting fast mode had almost no impact on the speed of output. Normal mode is already quite fast, and the code quality is the same, so it was a simple way to save money that we didn’t even notice,” Sartak noted.
However, the startuper believes that their “bottleneck” is the speed of releases, not the few thousand dollars per token. Slowing down to control token costs would cost them much more due to lost momentum.
“To save tokens, we use the simplest solutions: a standard default mode, lighter models for simple tasks, and not loading the entire codebase into the context. However, we do not worry about every dollar. We still adhere to this approach: if it makes us more productive, then in the long run it will most likely fully justify itself for the business,” Sartak concluded.
Як нейромережі бачать вільну та незалежну Україну? Тест dev.ua
Нейронні мережі для генерації зображень бачать світ по-своєму, їхню логіку зрозуміти часом зовсім неможливо. Але таки хочеться. На честь Дня Незалежності України редакція dev.ua вирішила провести невеликий експеримент.
Ми задали чотирьом різним нейронним мережам п’ять однакових запитів: «прапор України», «День Незалежності України», «український Крим», «перемога України» та «українці». Отриманими результатами ми ділимося з вами нижче.
У TikTok тепер можна генерувати фон за допомогою нейромережі. Ми протестували її та ділимося результатами
У TikTok з’явилася нова функція «Розумний фон». З її допомогою як фон для тіктоків можна підставляти згенеровані нейромережею зображення. Редакція dev.ua протестувала цю технологію і ділиться своїми враженнями.
