LLM API cost calculator for AI agents

Q: How can an agent know the price before paying?

GET https://pay.trustai.center/v1/quote?model= &input_tokens= &max_tokens= returns the worst-case USD cost, the exact x402 amount in USDC (6 decimals) and up to 3 cheaper alternatives. No key, never charges. Also available as the MCP tool quote.

Updated 1 November 2026 (live prices, same numbers as /v1/models).

Short answer: on the most-used open models an agent step (20k input + 1k output tokens) costs a fraction of a cent; 1,000 steps cost from about $1 to about $12 depending on the model (table below, live). Prices are at most 2% above the model maker's list price. Every call has a $0.001 minimum. Check any request before paying with GET /v1/quote (no key, never charges).

Cost per call, most-used models

Model id	$/1M in · out	Chat turn 2k in · 500 out	RAG answer 8k in · 800 out	Agent step 20k in · 1k out	1,000 agent steps
`deepseek-v4.1-flash`	$0.15 · $0.6	$0.0010	$0.0017	$0.0036	$3.60
`glm-5.3-flash`	$0.15 · $0.5	$0.0010	$0.0016	$0.0035	$3.50
`gpt-oss-120b`	$0.1 · $0.5	$0.0010	$0.0012	$0.0025	$2.50
`deepseek-v3.2-thinking`	$0.28 · $0.42	$0.0010	$0.0026	$0.0060	$6.02
`qwen3.7-flash`	$0.03 · $0.13	$0.0010	$0.0010	$0.0010	$1.00
`glm-4.7-flash`	$0.07 · $0.4	$0.0010	$0.0010	$0.0018	$1.80
`kimi-k2-thinking`	$0.47 · $2	$0.0019	$0.0054	$0.011	$11.40
`llama-4-scout`	$0.1 · $0.3	$0.0010	$0.0010	$0.0023	$2.30
`claude-haiku-5.5`	$0.1 · $0.5	$0.0010	$0.0012	$0.0025	$2.50
`gpt-5.6-luna`	$0.2 · $1.2	$0.0010	$0.0026	$0.0052	$5.20
`grok-4.1-fast-non-reasoning`	$0.2 · $0.5	$0.0010	$0.0020	$0.0045	$4.50
`mistral-nemo`	$0.15 · $0.15	$0.0010	$0.0013	$0.0032	$3.15

Billed amounts, computed with the same formula as billing: input tokens × input price + output tokens × output price, at least $0.001 per call. Prepaid keys pay the tokens actually used; x402 pays the worst case for max_tokens.

Calculator

Runs /v1/quote live.

Pick a model and press Quote.

Do it from an agent

curl "https://pay.trustai.center/v1/quote?model=deepseek-v4.1-flash&input_tokens=20000&max_tokens=1000"
# -> worst_case_usd, x402.amount_atomic (USDC, 6 decimals), cheaper_alternatives

Same thing over MCP: tool quote on https://pay.trustai.center/mcp. Or POST /v1/quote with your messages to price the exact prompt (upper-bound token estimate, same as the x402 price).

Tips to pay less

Set max_tokens close to what you need: x402 charges the worst case for it.
Use a prepaid tai_ key for many small calls: you pay tokens actually used, not the worst case.
Reasoning models bill hidden thinking tokens as output; use a non-thinking model for simple steps.
Check cheaper_alternatives in the quote.

FAQ

How can an agent know the price before paying?

GET /v1/quote?model=<id>&input_tokens=<n>&max_tokens=<n> returns the worst-case USD cost, the exact x402 amount in USDC and up to 3 cheaper alternatives. No key, never charges. Also available as the MCP tool quote.

Is there a minimum per call?

Yes, $0.001 per request. Small calls on cheap models are billed at that minimum.

What is the difference between x402 and a prepaid key?

x402 charges the worst case for max_tokens, settled only after a successful completion. A prepaid tai_ key is charged the tokens actually used (at least the minimum). Fund one with USDC or pay per call with x402.