LEA / cost lab

Open source · local first · budget native

Stop paying frontier prices for mechanical tokens.

LEA plans with capable models, implements with efficient or local models, and reviews with an independent provider—inside a budget you set before the run.

Explore the estimate Install LEA

No API key. No model call. The calculator uses the repository's versioned price catalog.

lea route

$ lea benchmark

  1. 01plangrok-4.6$0.0280
  2. 02codedeepseek-v4-flash$0.0042
  3. 03reviewclaude-sonnet-5$0.0460
  4. 04fixreserved review fix$0.0028
backend feature estimate$0.081
55.7%lower catalog estimate
4fixed coding scenarios
3+providers in a typical route
$0API cost to reproduce

Interactive catalog estimate

Put a route under your budget.

Choose a fixed scenario and a single-model baseline. Both sides use identical stage-level token assumptions. The budget counts affordable runs of this fixed route; it does not reroute the models.

$

Prices are USD per 1M tokens. Local Ollama costs are excluded because hardware and power are not comparable to hosted token prices.

All four benchmark scenarios
LEA route$0.2925
Grok 4.6 baseline$0.6596
55.7% lower estimated catalog cost
A $1.00 budget fits about 3 complete benchmark sets at this estimate.

What LEA routes—and why

Catalog checked 2026-09-04

Different optimization target

One agent. Multiple economic jobs.

01

Spend on leverage

Planning gets the reasoning budget because a good plan shapes every later token.

02

Save on volume

Implementation can move to an efficient hosted model—or Ollama with zero hosted API token cost.

03

Diversify review

Review prefers another provider, reducing the chance that one model grades its own blind spots.

04

Keep control live

Queue a follow-up or interrupt the HTTP call, retry wait, local command, or trusted MCP tool.

Illustrated feature walkthrough

See the budget before spending it.

Preview the exact stage route as JSON, reproduce the benchmark without credentials, then decide whether the run is worth its cap.

The animation is a drawn illustration, not a captured live run. Its 55.7% figure refers to the separate hosted-model benchmark above, not the illustrated local-model route.

lea route --budget 10cny --json "Add OAuth"
lea benchmark --json
lea run --budget 1usd "Ship the change"
LEA terminal walkthrough showing planning, coding, independent review, budget tracking, and live steering

Read the fine print

Useful estimate. Narrow claim.

This is a deterministic catalog-price estimate—not a claim that different models produce equal-quality work and not a prediction of your invoice.

  1. Classify four fixed tasks locally.
  2. Build LEA's default balanced route.
  3. Apply the same role-level token counts to both routes.
  4. Price them from the versioned provider catalog.
Read the full methodology

Budget the run. Route the work. Keep the steering wheel.

Try LEA on a version-controlled project.

Star and install on GitHub Shape the roadmap