Skip to Main Content
Economicsecon-067P1

Multi-turn conversations accumulate quadratic token costs.

A 10-turn conversation accumulates 15K…A 10-turn conversation accumulates 15K context tokens, costing $0.075 per session on GPT-4; conversation summarisation reduces this to $0.012 — an 84% saving.

Context & Methodology

Without conversation summarisation, each turn resends the full history, causing cost growth that scales quadratically with conversation length.

Applicable Use Cases

chat

Applies To

openaianthropicgoogle

Primary Impact

cost

Confidence Level

High

Platform Status

Built

Implementation Effort

medium

Recommendation

follow

Execution Priority

P1

Dependencies & Conflicts

Depends on:

Put This Evidence to Work

Use the STCO framework to implement findings like this in structured, testable prompts.

Deploying weekly canary prompts with known outputs detects performance regression with 95% accuracy within 24 hours of p.LangSmith, 'Evaluations and Monitoring' documentat…