Multi-Model Router · for AI engineers & FinOps
Route simple queries to Haiku. Keep Opus for the hard ones.
Most production workloads don't need a frontier model for every query. A 70/25/5 split between cheap/mid/premium typically saves 40-70% with no quality loss - provided you classify correctly. See exactly what you'd save.
Pricing verified: 2026-06-05
106 models
Classifier overhead included
What this calculator does
Route easy queries to Haiku 4.5 / GPT-5-mini / Gemini 3.1 Flash; reserve Opus 4.8 / GPT-5.5 / Gemini 3.1 Pro for the hard ones. See exactly how much you save.
Why use it
- Typical production workloads see 40-70% cost reduction — no quality loss if routed correctly
- Pick from 6 preset mixes (RAG, support, coding, research, content, balanced) or build your own
- Includes classifier overhead — LLM-as-judge / embedding / none — because the classifier itself costs something
- Compare your mix against every preset + sanity-check extremes (all-cheap, all-premium)
These are the inputs, outputs, and how you can use this calculator for your AI workloads.
📥 Inputs you provide
- Monthly requestsTotal monthly call volume
- Input tokens / requestAverage input size
- Output tokens / requestAverage output size
- Tier splitCheap / mid / premium query mix
- Baseline modelThe single model you compare against
- Routing methodHow queries get assigned to a tier
📤 Outputs you get
- Baseline costCost with one model for everything
- Routed costCost of your tier mix
- Monthly savingsDollars saved per month
- Classifier overheadCost of the routing decision
🎯 Use your results to
🔀
Tune your tier split
Test cheap/mid/premium ratios; many workloads hit 40-60% savings around an 80/15/5 mix
🆚
Compare against baseline
Exact dollar gap vs running one model for everything, plus every preset and the extremes
⚙️
Price the classifier
Routing overhead netted out so the savings are honest, not gross
🔌
Integrate with your agents
MCP available so agentic workflows can pull routing economics programmatically
👇 Now try the calculator below with your own AI workloads
1
Enter every field
Monthly requests drives absolute scale. Input/output tokens per request affect each tier equally — tune to your real average. Then the 3-tier split: cheap % (simple queries — FAQ, retrieval Q&A, autocomplete), mid % (standard reasoning — explanations, synthesis), premium % (hard reasoning — architecture, edge cases). Sum should equal 100; sliders auto-adjust. Baseline selector = the single model you'd use if you weren't routing (defaults to your premium tier).
2
Pick a classifier strategy
No classifier = rules-based (regex, length, intent-in-URL) at zero overhead. LLM-as-judge = small model classifies each query at ~$0.0001/query (essentially free at scale). Embedding = embed query, match to intent clusters — cheaper still but ~85-92% accuracy vs ~90-95% for LLM-as-judge. Mis-routed queries trigger retries and can erode savings; factor in a 5-10% "escalation overhead" if accuracy is tight.
3
Interpret all panels
Top 3 cards compare baseline vs routed. Per-tier blocks show monthly cost for each tier + per-request cost — use this to spot tiers where you're paying more than necessary. Cost-share bar visualizes routed dollar distribution including classifier. Comparison table stacks your mix against every preset AND sanity-check extremes ("all cheap", "all premium") — useful for bounding max savings.
4
Next actions
Savings should be validated before shipping. Run offline eval: does your cheap-tier model handle your cheap-tier queries at acceptable quality? If yes, deploy with monitoring. Monitor the cheap-tier escalation rate (how often cheap-tier output is rejected and retried on premium) — if >10%, your classifier needs tuning. Stack with prompt cache and batch API — savings multiply, not add.
📊 Calculator at a glance
🎛 CALCULATOR
🧭 Your workload shape
Start with a preset, then tune the mix.
⚠️ Your tier percentages don't sum to 100% - results will be normalized.
🟢 Cheap tier - simple queriesHow your traffic divides across cheap, mid, and premium models. This is the core routing decision — most production workloads have 50-70% of queries a cheap model handles fine.How to choose: Cheap (🟢): simple/structured work — classification, extraction, FAQ, autocomplete, status checks. Mid (🟡): standard reasoning — explanations, synthesis, summaries. Premium (🟣): hard reasoning — architecture, multi-step logic, edge cases. Start from a preset, then drag the sliders; they co-adjust toward 100%.Read the full guide →
70%
-
🟡 Mid tier - standard queries
25%
-
🟣 Premium tier - complex queries
What would you be using if you weren't routing? Baseline defaults to your premium-tier model.
-
📈 RESULTS
Baseline (single model)
-
-
With routing
-
-
Monthly savings
-
-
💡 Recommendations
📋 Compare routing strategies side-by-side
All strategies at your workload. Green = biggest savings, gold = your current mix.
| Strategy | Mix | Monthly | vs baseline | Savings |
|---|
📋 What now?
- Validate the cheap tier before you ship — run an offline eval on your simple-query slice; the savings only hold if the cheap model answers those at acceptable quality.
- Add an escalation path — route low-confidence answers up to premium so misroutes don't turn into retries and complaints. Watch the cheap-tier escalation rate; above ~10% means the classifier needs tuning.
- Then stack the other levers — prompt caching and batch processing multiply with routing rather than just adding, often another 20-40% off the routed number.