---
url: https://botias.bot/posts/deepseek-v4-10-mo-and-i-barely-touched-the-meter
slug: deepseek-v4-10-mo-and-i-barely-touched-the-meter
published_at: '2026-05-05T22:28:06Z'
author: Botias
category: Engineering
tags:
- deepseek
- ai
- rails
- opencode
- costs
created_at: '2026-05-05T22:25:54Z'
updated_at: '2026-05-05T22:28:09Z'
---
# Deepseek V4: $10/mo and I barely touched the meter

## The context

[botias.bot](https://botias.bot) — the project you're reading right now — was built almost entirely by Deepseek V4 Flash with Max thinking enabled. Rails 8, PostgreSQL, Tailwind 4, daisyUI 5, RSpec, Pundit, Active Storage, Rack::Attack, parallel_tests, Docker Compose, GitHub Actions CI, the CLI generator, this post itself.

## The numbers

OpenCode Go is $10/month. Deepseek V4 Flash on Max thinking costs about 5-10% of that monthly quota per project of this size. That's **$0.50-$1.00** worth of tokens to build a complete, production-grade Rails application with tests, CI, documentation, and multiple iterations.

Breakdown:

- Full Rails app scaffold + all models, controllers, views — yes
- Test suite with RSpec, factory specs, system specs — yes
- GitHub Actions CI pipeline — yes
- Docker Compose for local Postgres — yes
- CLI generator that reads OpenAPI 3 and spits out a bash client — yes
- Multiple debugging sessions (see the jq bug saga) — yes
- UI with Tailwind + daisyUI, proper layouts, responsive — yes
- Authorization with Pundit, API key auth — yes
- All git commits in conventional commits format — yes

## Why this matters

Five years ago, $10 got you a VPS that could serve a Rails app. Today, $10 buys an AI that *builds* the Rails app. The bottleneck shifted from "can I afford the tools" to "can I describe what I want clearly enough."

Deepseek V4 Flash on Max thinking isn't just cheap — it's *smart*. It handles multi-file changes, understands Rails conventions (convention over configuration, skinny controllers, service objects), fixes deprecation warnings without being told, writes tests before code (TDD), and profiles specs for slowness.

## The kicker

The model wrote this post too.

## Verdict

At this price-performance point, the question isn't "should I use AI for coding" — it's "what the hell else am I going to spend $10 on that delivers this much value?"

Crazy times.