How much does the GPT, Claude & Gemini API actually cost?
Short answer: you pay per token, not per request — and the range is wide. Budget models cost cents per million tokens; premium models cost several dollars per million, with output priced 4–5× higher than input. Your actual bill is total tokens × per-token price, summed across every call. Below are the current rates by tier, a worked example, and the three things that really drive the number.
Want your own figure? The cost calculator does the math from your usage. Otherwise, the rates:
Per-token rates by tier (2026)
| Model | $ / 1M in | $ / 1M out |
|---|---|---|
| Gemini-class flash budget | 0.10 | 0.40 |
| GPT-class mini budget | 0.15 | 0.60 |
| Claude Haiku 4.5 budget | 1.00 | 5.00 |
| Gemini-class pro mid | 1.25 | 5.00 |
| GPT-class frontier mid/prem | 2.50 | 10.00 |
| Claude Sonnet 5 mid | 3.00 | 15.00 |
| Claude Opus 4.8 premium | 5.00 | 25.00 |
The headline pattern: a budget model can be 10–25× cheaper than a premium one, and output always costs several times more than input. Which tier you use is the biggest cost decision you make — more than which vendor. See Claude vs GPT vs Gemini for a workload-by-workload comparison.
A worked example
Say you run 1,000,000 requests/month, each with ~1,000 input and ~500 output tokens. That's 1,000M input and 500M output tokens/month. Cost on a few tiers:
| Model | Monthly cost |
|---|---|
| Budget (mini/flash class) | ~$150–450 |
| Claude Haiku 4.5 | $3,500 |
| Claude Sonnet 5 | $10,500 |
| Claude Opus 4.8 | $17,500 |
Same traffic, a 40–100× cost swing — entirely down to model choice. That gap is why routing easy work to a cheaper model is the single biggest lever on an LLM bill.
Get your exact number
Plug in your requests and token sizes; edit any price to match today's rates; see every model ranked cheapest-first.
Open the calculator →What actually drives your bill
1. Total tokens, not requests
You're billed on tokens, so volume × tokens-per-request is what matters. A million small calls can cost less than a hundred thousand context-stuffed ones. Reducing tokens per request (tighter prompts, leaner retrieval) lowers cost directly.
2. Output vs input
Output is ~4–5× the input price. Long-answer workloads (drafting, code gen) are dominated by output cost; retrieval-heavy workloads by input cost. Capping max_tokens controls the expensive side.
3. Discounts you may be ignoring
Prompt caching (~90% off repeated prefix), the Batch API (50% off non-urgent work), and smaller models can each cut a bill substantially. The full list is in how to cut your bill 50%+.
FAQ
How much does the GPT API cost?
It depends on tier and volume — budget models are cents per million tokens, premium models several dollars per million input and 4–5× that for output. It's per token, not per request.
Is the Claude API expensive?
Claude spans tiers: Haiku ($1/$5) is budget, Sonnet is mid, Opus ($5/$25) is premium. The tier drives cost far more than the vendor.
What drives my LLM bill?
Total tokens, with output priced ~4–5× higher than input. Context-heavy and long-output requests cost the most.