Skip to content

Cost-Quality Pareto Frontier

Run the MicroSim Fullscreen

About This MicroSim

A scatter plot of nine representative model configurations on log-scale cost vs quality. The Pareto frontier (green) connects non-dominated points — those for which no other configuration is both cheaper and higher quality. Dominated points (gray) should never be chosen. Sliders apply quality-floor and cost-ceiling constraints, gray-fading any configuration that fails the test, leaving only the survivors.

How to Use

  1. Read the green frontier. These are the only configurations worth considering. The dashed line connects them.
  2. Notice the dominated points. Any gray point — its cost-quality position is strictly worse than some green point. Never pick these.
  3. Set quality floor to 80. Watch some frontier points fade. Read the survivor list at the bottom.
  4. Set cost ceiling to $0.01. Combined with quality 80, the survivor set narrows further. Often it collapses to a single configuration — the right pick.

Bloom Level

Evaluate (L5) — judge which model configurations are worth considering for a given workload and which are strictly dominated.

Iframe Embed Code

1
<iframe src="sims/cost-quality-pareto-frontier/main.html" height="582px" width="100%" scrolling="no"></iframe>

Lesson Plan

Audience

Engineers and ML engineers selecting model configurations for production workloads.

Duration

10–15 minutes inside Chapter 3.

Prerequisites

Chapter 3 sections on Cost-Quality Tradeoff, Pareto Frontier, Cached Input Price.

Activities

  1. Identify the frontier (3 min). Without sliders, name every green point.
  2. Spot the dominators (3 min). For each gray point, name the green point that strictly dominates it and explain why.
  3. Apply constraints (5 min). Set quality floor to 85, cost ceiling to $0.05. List the survivors. Argue for the best pick.

Practice Scenarios

# Quality floor Cost ceiling Survivor
1 0 $1.00 (full frontier)
2 80 $0.05 ?
3 90 $0.20 ?
4 75 $0.005 ?
5 95 $1.00 ?

Assessment

Learner has met the objective when, given a new configuration, they can decide whether it joins the frontier or is dominated.

References

  1. Multiple-Criteria Decision Analysis — for the formal Pareto-dominance definition.
  2. Anthropic / OpenAI / Google pricing pages — for the actual configurations plotted here.
  3. Chapter 3 of this textbook — Cost-Quality Tradeoff.

Senior Instructional Designer Quality Review

Reviewer perspective: 15+ years designing engineering and decision-science curricula for adult professional learners.

Overall verdict

Approve as-is for Chapter 3. Score: 89/100 (B+). Pareto-frontier visualization is the canonical primitive for L5 "judge"; this implementation makes the dominated set undeniably visible.

What works

  1. Bloom alignment correct. L5 "judge" requires applying explicit criteria; the constraint sliders externalize the criteria.
  2. Log-scale X axis. Without it, the cheap-model cluster vs expensive-thinking cluster wouldn't both be visible.
  3. Dashed frontier line. Visually proves the "non-dominated" claim — every green point is reachable on the line.
  4. Survivor list in the status banner. Translates filtering to action.

Gaps

  1. Only 9 configurations. Real teams have many more (different prompts, caching variants, vendor combos). A "load my own configurations" or "add a custom point" affordance would teach the methodology, not just the answer. Score impact: −3.
  2. No way to define a custom quality metric. Different workloads care about different quality dimensions; the single 0–100 score conflates them. Score impact: −2.
  3. No explicit dominance explainer. Hovering a gray point should show "dominated by: " in the tooltip. Score impact: −2.

Accessibility

Color-blind safe (green/gray with size differences). Tooltips have full text. Slider labels show numeric values.

Cognitive load

9 points + 2 sliders + frontier line. Tractable.

Recommendation

Approve. Open follow-up for hover-explains-dominance (gap 3) — small but high-impact addition.