Last month, I shared an analysis breaking down the costs of talking to AI models 24/7 for a month, comparing OpenAI's most expensive tier with Together AI's budget-friendly option. The response to that post was amazing, and given the buzz around DeepSeek, I wanted to make a new comparison between OpenAI’s & DeepSeek’s most advanced models.

Revisiting the Numbers

For consistency, I used the same assumptions from my last analysis:

• Conversation Speed: 60 words per minute.

• Word-to-Token Ratio: 1 word equals 1.3 tokens.

• Continuous Interaction: 24 hours a day, every day.

This means a daily token count of: 60 words/min × 60 min/hr × 24 hr/day × 1.3 tokens/word = 112,320 tokens/day.

Monthly token count: 112,320 tokens/day × 30 days = 3,369,600 tokens/month.

Let’s crunch the numbers with DeepSeek’s pricing tiers and OpenAI’s most advanced model.

1. DeepSeek’s Pricing

DeepSeek offers the Reasoner model:

• DeepSeek-Reasoner:

   • Input tokens: $0.55 per 1M tokens.

   • Output tokens: $2.19 per 1M tokens.

For simplicity, I’ll assume a 3:1 ratio of input to output tokens, a common pattern in AI usage.

Cost Breakdown for DeepSeek-Reasoner:

• Input Tokens: 3,369,600 tokens × (3/4) = 2,527,200 tokens.

• Output Tokens: 3,369,600 tokens × (1/4) = 842,400 tokens.

• Input Cost: (2,527,200 / 1,000,000) × 0.55 = $1.39

• Output Cost: (842,400 / 1,000,000) × 2.19 = $1.85

Total Cost for DeepSeek-Reasoner: $1.39 + $1.85 = $3.24/month.

2. OpenAI’s Most Advanced Model

OpenAI’s most advanced model, o1, offers the following pricing:

• Input tokens: $15.00 per 1M tokens.

• Cached input tokens: $7.50 per 1M tokens.

• Output tokens: $60.00 per 1M tokens.

Cost Breakdown for OpenAI o1:

• Input Tokens: 3,369,600 tokens × (3/4) = 2,527,200 tokens.

• Output tokens:  3,369,600 tokens × (1/4) = 842,400 tokens.

• Input Cost: (2,527,200 / 1,000,000) × 15.00 = $37.91

• Output Cost: (842,400 / 1,000,000) × 60.00 = $50.54

Total Cost for OpenAI o1: $37.91 + $50.54 = $88.45/month.

Observations

• 27x price difference;

• OpenAI updated its pricing since last month. In the last post, the estimated cost was about $404 / month. So 125x price difference here.

While DeepSeek is the obvious choice from a cost perspective (probably performance too, but I guess still to be validated), I wonder how strong the brand value of OpenAI or other big providers actually is. I have doubts companies will start deploying from a new Chinese vendor right away due to so many reasons. We still have to see if DeepSeek will be able to cross this skepticism bridge.

Closing Thoughts

In my previous post I stated we’re in a race to the bottom. Well, it’s still going on hard! I guess this was pretty foreseeable, but soon enough we’ll wonder how sustainable this is. I believe costs will continue to decrease

In this case, will startups still need to stress about ways to adapt to this new COGS line in their P&L (e.g. founders investing in their own mini data centers)?