What

achieve #131
Added
1.cost-tiered-pipeline.ts
a new patterns/cost-tiered-pipeline.ts example that runs the same four-stage runTasks() pipeline twice:

1. A flagship-only baseline
2. A tiered model mix

The example captures token usage via onTrace, estimates USD cost from a dated PRICING table, prints a side-by-side comparison table,
and asserts that the tiered run achieves at least 40% cost savings.

Also add a one-line entry for the example in examples/README.md.

Prerequisites

OPENAI_API_KEY='your_KEY'
OPENAI_BASE_URL='your_URL'
LLM_PROVIDER='openai'

Why

Fixes #131

This makes the cost story behind per-agent model assignment concrete with a runnable example, instead of leaving it as an abstract framework capability.

Checklist

• npm run lint passes
• npm test passes
• Added/updated tests for changed behavior (not needed; example-only change)
• No new runtime dependencies

Notes

• The example supports both anthropic and openai-compatible provider paths.
• Taking into account model performance and input/output pricing, the counterpart gpt following model has been selected.(date 26.4.24)
  'gpt-5.5': { input: 5, output: 30 },
  'gpt-5.4': { input: 2.5, output: 15 },
  'gpt-5.4-mini': { input: 0.75, output: 4.5 },
• Verified locally with an OpenAI-compatible endpoint and a GPT tiered mix.

Local run achieved 55.1% cost savings versus the baseline,
Models: researcher=gpt-5.5, classifier=gpt-5.5, drafter=gpt-5.5, reviewer=gpt-5.5
Tokens — input: 6229, output: 2606
Estimated USD cost — $0.1093
Wall time — 56.02s
Models: researcher=gpt-5.4-mini, classifier=gpt-5.4-mini, drafter=gpt-5.4, reviewer=gpt-5.5
Tokens — input: 5914, output: 2199
Estimated USD cost — $0.0491
Wall time — 62.88s