70% savings claimed, but what’s the actual math? ChatGPT Team is $25/user/month, so you're comparing a flat subscription to a per-token API.
If your “acceptable quality dip” means you’re using DeepSeek’s cheaper models for simpler tasks, fine. But for parity (ChatGPT-4o-level work), you’re likely using DeepSeek-V2. Let’s run numbers:
- ChatGPT Team: $25/month, unlimited messages (within fair use).
- DeepSeek-V2 via API: $0.14 per 1M input tokens, $0.28 per 1M output tokens.
Assume your team’s monthly usage:
* 500,000 messages (light)
* Avg 500 tokens per message (input+output split 50/50)
* Total monthly tokens: 250M
* Cost = (125M input * $0.14/1M) + (125M output * $0.28/1M) = $17.50 + $35.00 = $52.50
You’re already **double** the ChatGPT Team cost. To hit a 70% cut, your usage must be tiny or you’re using a much weaker model.
Show me:
* Your actual monthly token consumption
* Which DeepSeek model you’re calling
* What “quality dip” tasks you offloaded
Otherwise, this is just another “I saved money by doing less” post.
— show the math
show the math
Good point, the math does look rough for higher usage. But what if they're a smaller team? If they only have 3 users and maybe 50k messages a month, the API cost could be way under $75 for the Team plan total. That's where the 70% might come from.
Also, is the "fair use" on ChatGPT Team actually unlimited? I've heard about rate limits that could make high-volume work tricky. Maybe they hit those and needed the API for scaling, even at a higher cost per token?
What do you consider "tiny" usage for this to make sense? Trying to learn for my own team's evaluation