Cost Optimization

How RelayPlane saves you money without sacrificing quality.

The Problem

AI coding agents send everything to expensive models. A simple "list files in this directory" costs the same as "architect a distributed system." Most tasks don't need Opus-level reasoning.

Without RelayPlane
  • Every request → Opus ($15/MTok)
  • Tool calls → Opus
  • Simple questions → Opus
  • ~$100-300/month typical
With RelayPlane
  • All requests → Haiku ($0.25/MTok)
  • 60x cheaper per token
  • Same API, same tools
  • ~$5-20/month typical

How Routing Works

By default, RelayPlane routes all requests to Haiku for maximum cost savings. The proxy tracks task types to help you understand your usage patterns:

Task TypeDefault ModelCost
tool_useHaiku$0.25/MTok
quick_taskHaiku$0.25/MTok
code_reviewHaiku$0.25/MTok
generationHaiku$0.25/MTok
long_contextHaiku$0.25/MTok
generalHaiku$0.25/MTok
Why Haiku for everything? Most agent tasks (tool calls, file reads, simple edits) don't need Opus-level reasoning. The proxy optimizes for cost by default. Quality routing with model escalation is coming in a future release.

Typical Results

40-60%
Cost Reduction
~70%
Tasks Routed Down
0ms
Added Latency

View Your Savings

1relayplane-proxy stats

Example output:

1📊 Usage Statistics
2═══════════════════
3
4 Total requests: 1,247
5 Total cost: $42.18
6 Success rate: 95.3%
7
8 By Model:
9 claude-3-5-haiku: 847 requests, $8.12
10 claude-sonnet-4: 312 requests, $24.06
11 claude-opus-4: 88 requests, $10.00
12
13 By Task Type:
14 tool_use: 523 requests, $5.00
15 quick_task: 324 requests, $3.12
16 code_review: 248 requests, $24.06
17 generation: 152 requests, $10.00
No Quality Loss: Complex tasks still go to Opus. You only save money on tasks that don't need premium reasoning.

Real-World Example

A typical OpenClaw coding session might look like:

  • 50 tool calls (read file, list dir, search) → Haiku
  • 15 code reviews/edits → Sonnet
  • 3 architecture decisions → Opus

Without RelayPlane: All 68 requests to Opus = ~$15
With RelayPlane: Mixed routing = ~$6