Token Cost
Token cost is the price you pay an LLM provider per unit of input or output text — measured in dollars per million tokens, with input usually cheaper than output.
Tokens are roughly 0.75 words. Frontier models in 2026 cost from $0.30 (Haiku-tier) to $15+ (Opus-tier) per million output tokens. Token cost compounds fast in agentic workflows: a single mission can spend hundreds of thousands of tokens across planning, retrieval, tool calls, and output. The right comparison is always cost per task, not cost per token.
Example
An AI SDR drafts a 200-token email after reading 4,000 tokens of context. At $3/M input + $15/M output, the draft costs ~$0.015. At 1,200 emails/week that's ~$18/week.
How OpenLabor uses it
OpenLabor surfaces token spend per employee per mission so you see where the money goes — and routes to cheaper models where quality allows.
Why does my token bill grow faster than my usage?
Usually retrieval + retries — the visible work is the tip; the iceberg is the context the model reads.
Related: cost-per-task, llm, context-window.
AI Labor Glossary