How much will GPT-4o / Claude cost at scale?
Select model
OpenAI
Anthropic
Groq / Together
Mistral
How much you charge above API cost
GPT-4o — 800 tokens avg
Cost per request
$0.00575
Cost per 1,000 requests
$5.750
Daily (1,000 reqs)
$5.75
Monthly
$172.50
Yearly
$2099
| Model | $/req | $/mo (1000/day) | Context |
|---|---|---|---|
| GPT-4o OpenAI | $0.00575 | $172.50 | 128K |
| GPT-4o mini OpenAI | $0.00034 | $10.35 | 128K |
| GPT-4.1 OpenAI | $0.00460 | $138.00 | 1000K |
| o3 OpenAI | $0.02300 | $690.00 | 200K |
| o4-mini OpenAI | $0.00253 | $75.90 | 200K |
| Claude Opus 4 Anthropic | $0.04200 | $1260.00 | 200K |
| Claude Sonnet 4 Anthropic | $0.00840 | $252.00 | 200K |
| Claude Haiku 3.5 Anthropic | $0.00224 | $67.20 | 200K |
| Gemini 2.5 Pro Google | $0.00538 | $161.25 | 1000K |
| Gemini 2.5 Flash Google | $0.00034 | $10.35 | 1000K |
| Llama 3.3 70B Groq / Together | $0.00057 | $17.16 | 128K |
| Mistral Large Mistral | $0.00360 | $108.00 | 128K |
Input vs output tokens
Input tokens = everything in the prompt (system prompt + user message + context). Output tokens = the model's response. Output is always 3–5× more expensive than input.
Cost optimization
Switch to GPT-4o mini or Gemini 2.5 Flash for simple tasks — they're 10–20× cheaper and 90% as good. Use expensive models only for complex reasoning or long outputs.
Pricing your AI SaaS
Cover API cost + 200–500% margin. If GPT-4o costs you $0.10/request, charge $0.30–$0.50. Build usage limits per plan to protect your margin.
RAG cost optimization
RAG input costs are high (lots of context). Cache repeated context with OpenAI's prompt caching ($0.5/1M cached tokens vs $2.5/1M uncached). 80% savings for repeated system prompts.
We ship production-ready MVPs in 15–21 days at a fixed price. No surprises, no hourly billing.