Independent reference. We are independent of every vendor listed. No affiliate links. No sponsored placements.
Topic guide

Platform Cost Is Half the Bill

The LLM token line on an agentic deployment often equals or exceeds the platform list price. Forecast both.

The two lines

$3-15
Per 1M tokens
Anthropic Claude Sonnet input / output
$5-30
Per 1M tokens
OpenAI GPT-5.5 input / output
$1.50-9
Per 1M tokens
Google Gemini 3.5 Flash input / output
60/40
Typical mix
Input tokens / output tokens on a long agent run

Worked example

Illustrative example, not a real company. 10,000 agent interactions per month, 12,000 tokens average per interaction at a 60/40 input/output mix on Claude Sonnet.

Token line (Claude Sonnet)
0110,000 interactions x 12,000 tokens
120M tokens
02Input 60% x $3 / 1M = 72M x $3
$216
03Output 40% x $15 / 1M = 48M x $15
$720
Monthly token line$936
Platform line (illustrative)
01If on Agentforce $2 / conversation
$20,000
02If on Devin Pro seat
$20
03If on LangGraph Plus + LCU
$200-$500
Platform varies wildly by model$20 to $20,000

The token line is roughly $936 regardless of which platform you pick (the model does the work either way). The platform line varies by 1000x. This is the gap that procurement papers ignore.

The grounding line

Third line on a real bill

Beyond platform and token, grounded agents add a third line: search / retrieval calls. Vertex AI bills $14 per 1,000 grounded Google Search queries above 5,000 free per month from January 2026, plus $2.50 per 1,000 prompts for grounding with your own data via Vertex Search. Anthropic and OpenAI grounding via external retrieval is billed by the retrieval vendor (Tavily, Exa, Brave Search API).

How to forecast both lines

The platform line is your monthly subscription. The token line is your monthly electricity bill. Vendor pricing pages show you the rent; they do not show you the electricity. Forecast both.
AgenticAICost.com editorial

Continue with Agent TCO for the full year-one cost stack.

Last verified June 2026