June 26, 2026
Token shock hits the chatbots
Why current LLM costs are not sustainable
AI’s pricey chatbot era may be ending as users roast the bill and hunt for cheaper options
TLDR: Advanced AI tools are amazingly useful, but their current prices may not last as cheaper rivals and better hardware drive costs down. In the comments, readers turned that into a fight over layoffs, privacy, subscriptions, and whether the whole industry is headed for a very expensive reality check.
The big vibe in the comments is basically: the AI party bill has arrived, and nobody wants to split it. The article argues that today’s most advanced chatbots are simply too expensive to stay this pricey forever, especially as improvements get smaller, cheaper rivals show up, and companies find they can switch tools with almost no pain. The moment that really set people off? The author saying a single afternoon of fixing coding errors across 50 files cost $54. That number hit the thread like a jump scare.
From there, the crowd split into camps. One side says the future is obvious: businesses will stop paying luxury prices and move to cheaper, more private tools they can run themselves. Commenters were already doing back-of-the-envelope math on shared in-house setups, with one person loving that your data and chat history would stay private and couldn’t be switched off by some outside company. The other side was more cynical: forget replacing the expensive tools entirely, some firms will just cut staff and turn salaries into “token budget” instead.
There was also plenty of delicious snark. One commenter basically said, why pay by the word when the big companies are heavily subsidizing monthly subscriptions anyway? Another shrugged at the possible fallout for giant AI firms, joking that the “disaster” would simply be them dropping back to a still-ridiculous $5–10 billion valuation. In other words: the technology is exciting, but the comments section is already planning the clearance sale.
Key Points
- •The article says enterprise AI costs are rising quickly, citing Uber’s reported budget burn and cost-control measures at Microsoft, Salesforce, and GitHub.
- •It presents GPT 5.5 pricing and a personal TypeScript example to show how frontier-model usage can become expensive in routine workflows.
- •The article argues that model performance gains are plateauing, which may make it harder for labs to justify continued high prices.
- •It says open-weight models such as GLM-5.2 and broad inference-provider access can undercut frontier labs on price.
- •The article identifies AI-specific chips, improved model architectures, and low switching costs as factors that could push LLM prices down over time.