LLM Context Window Cost Comparison
Long-context calls are the most expensive shape of agent spend — input tokens dominate when the context is big. Here's the honest math.
Why context dominates cost
Input tokens are billed per token; a 100K-token context costs 100× a 1K context before a single output token.
Agent tool outputs, documents, and conversation history all pile into context.
The rough math
At typical input rates, a single 100K-token call costs roughly the same as 50–200 short calls, depending on provider and model.
Cache-hit discounts change the picture dramatically for repeated contexts — see the cache savings benchmark.
What it means for agents
Pin context lengths: trim tool outputs, summarize history, cache stable prefixes.
Cap per-call spend so one oversized context can't blow the budget.
At a glance
| Context size | Relative input cost | Agent impact |
|---|---|---|
| 1K tokens | Baseline | Typical tool call |
| 10K tokens | ~10× | Conversation with history |
| 100K tokens | ~100× | Document-scale analysis |
| 1M tokens | ~1000× | Corpus-scale — expensive |
FAQ
Why is context the biggest lever?
Because input tokens scale linearly with context, and agents keep re-sending the same context. Trimming and caching are the fixes.
Are these exact prices?
No — rates change frequently. The shape (context × rate) is the point; verify current per-token prices with your provider.
How does sipi.bot help?
Per-transaction caps stop a single oversized call from blowing the budget, and category rules separate inference from other spend.
Related
Stop the next $12,400 night.
One API call (or MCP tool) in front of every agent transaction — APPROVED, BLOCKED, or FLAGGED, deterministic, ~5 ms, fully logged.
See plans — from $99/mo Try a live check