All LLMs I see in the current times offer two types of tiers: Free and Paid. Nothing in between. This made me wonder, why not ads? All the pre-LLM era products, where the user is not paying hard cash, ads have been the tried-and-tested model. So, why not now?
So I did some thinking, research and excel on that.
LLM companies incur these costs:
Now, all LLMs provide an absolutely free tier to the users. This, although limited in terms of number of of words (or tokens), works just fine for some users. Many users are willing to wait for the chat to resume at 11.57 pm (I am looking at you, Claude), if they don’t have to pay for it. But an LLM still has to pay for these free conversations. Free users are pure cost centers for LLMs. OpenAI loses $5 billion annually vs generating a $3.7B revenue - free users are pure cost centers.
Let’s calculate the cost incurred for a free user.
Assuming there would be 3 types of users:
Light (60% of free users) - looking for quick Q&A, short responses
Medium (30% of free users) - having longer discussions, follow-up questions
Heavy (10% of free users) - doing deep work sessions, coding, analysis in multiple context windows
| User type | Number of words in conversation with LLM (30% prompt+ 70% response) per day | Number of tokens consumed per day (30% input+ 70% output) | Token cost incurred per day |
|---|---|---|---|
| Light | ~1,500-4,500 | ~1,200-3,600 | $0.00155 - $0.00464 |
| Medim | ~4,600-19,20 | ~3,500-14,400 | $0.00452 - $0.01858 |
| Heavy | ~14,000-38,000 | ~10,500-28,500 | $0.01355 - $0.03677 |
Note: Cost of tokens is a vast range depending on the LLM and the model used. It ranges from $0.5-$3 per 1M input tokens and $5-$15 per 1M output tokens. Most used models by free users would cost on the lower range, hence for calculation, I have assumed cost of Input tokens as $0.8 per 1M tokens and Output tokens: $1.5 per 1M tokens. Since we have already assumed 30% as input and 70% output token distribution, the weighted average token cost is $1.29.
Let’s do a weighted average cost per free user to arrive at a single cost number, considering upper cost limit:
Light users: 60% × $0.00464 = $0.002784
Medium users: 30% × $0.01858 = $0.005574
Heavy users: 10% × $0.03677 = $0.003677