Cheaper Claude Code — pay per turn, or use a budget model

Steps

Step 1

$ curl -fsSL https://3-229-227-242.sslip.io/install.sh | sh

Install the CLI.

Step 2

$ aethene login

Step 3

$ aethene install claude --model aethene/gpt-6-luna

Make the budget model your default; switch to Opus with /model when a task needs it.

What happens

Caching is passed through

Claude Code resends your whole conversation each turn. Aethene passes prompt caching through, so most of those tokens bill at the cache-read rate — that's what keeps a typical turn near five cents.

A budget model on the same key

aethene/gpt-6-luna costs a small fraction of Opus per token. Use it for small edits and tests; keep Opus for multi-file work.

Every request itemised

Tokens, cost and plan units per request in the dashboard and `aethene logs`, so you can see exactly where the money goes.

As of 2026-09-25: a typical Claude Code turn costs $0.068 on aethene/claude-opus and $0.0020 on aethene/gpt-6-luna.

Models

Every model on the same key

Claude Opus 5.5

Anthropic's Opus flagship. Claude Code default.

Input: $5.00
Output: $25.00
Context: 1M

aethene/claude-opus≈ $0.068 / turn

GPT-6 Astra

OpenAI's frontier model for the most demanding work.

Input: $12.50
Output: $62.50
Context: 1M

aethene/gpt-6-astra≈ $0.197 / turn

GPT-6 Sol

Fast, capable all-rounder. Chat default.

Input: $2.50
Output: $12.50
Context: 1.1M

aethene/gpt-6-sol≈ $0.039 / turn

GPT-6 Luna

The budget model for quick, high-volume work.

Input: $0.13
Output: $0.63
Context: 1.1M

aethene/gpt-6-luna≈ $0.002 / turn

GPT-4o

The reliable multimodal classic.

Input: $3.13
Output: $12.50
Context: 128K

aethene/gpt-4o≈ $0.098 / turn

DeepSeek V4 Pro

DeepSeek's flagship. Strong coding and reasoning at a low price.

Input: $2.18
Output: $4.35
Context: 1M

aethene/deepseek-v4-pro≈ $0.023 / turn

DeepSeek V4.1 Flash

Fast, very cheap DeepSeek model for everyday coding.

Input: $0.55
Output: $2.74
Context: 1M

aethene/deepseek-v4-flash≈ $0.009 / turn

Gemini 3.8 Flash

Google's fast thinking model, straight from Google.

Input: $0.94
Output: $4.69
Context: 1M

aethene/gemini-3-8-flash≈ $0.014 / turn

Auto

Tools or long context go to Claude Opus; short chat goes to the cheapest healthy GPT-6. Billed at the model it picks.

aethene/autoCompare all →

Questions

FAQs

Is this cheaper than Claude Max?

Not for heavy all-day use — a Max subscription usually wins there. For light and medium use, teams and multi-model work, pay-per-turn is usually cheaper. The calculator shows your case.

How do I see what I'm spending?

aethene logs --last 20 in the terminal, or the dashboard's request log with CSV export. Every request shows tokens, cost and plan units.

Set up in three minutes.