CIMC · 12
0/10 done
MasterClass 12 · Crafted Intelligences

Code without expensive subs.

A Crafted Intelligences guide to OpenCode Go. Ten calm steps, about twenty minutes.

Organized by
Sam Alemayehu
Crafted Intelligences
Special guest
Ariya Hidayat
MC · 12
last reviewed · July 21, 2026
Chapter 00

The idea

00 – why this exists

You don't need a frontier sub to code well.

Frontier assistants are excellent – and expensive. For most Crafted Intelligences members, the simplest alternative is OpenCode Go: an open-source agent plus one capped subscription that unlocks a curated set of very good coding models.

No local server to babysit. No provider spreadsheet. One price, many models, and the ability to switch on purpose whenever the task changes.

step · You

You type a request in plain English – no special syntax.

  • "refactor this file"
  • "write tests for utils.ts"
  • "explain this error"
hover or tap a step · one subscription, many coding models
Optional side branch: OpenCode Zen – pay-as-you-go top-up for models Go doesn't include. Add later if you outgrow the plan.
Chapter 00.25

Meet the models

00.25 – the models on go

One sub. many good models.

These are the models included in an OpenCode Go plan, filtered by what you care about. Frontier models (Claude, GPT, Gemini) are shown below – they are optional and not part of Go.

included in opencode go · recommended

GLM 5.2

Zhipu

Everyday driver. Fast, calm, cheap – the model you'll reach for most days.

on goalso on zencodingfastcheapopen

Kimi K3 Code

Moonshot

Heavy reasoning. Reach for K3 on architecture, tricky debugging, or whole-repo work.

on goalso on zencodingreasoning

Kimi K2.7 Code

Moonshot

Value coder. Nearly K3 quality at a fraction of the cost – great as an alternate daily.

on goalso on zencodingcheap

DeepSeek V4 Pro

DeepSeek

Open weights, frontier-adjacent quality. Solid alternative when Kimi is busy.

on goalso on zencodingreasoningcheapopen

DeepSeek V4 Flash

DeepSeek

Simple, repetitive work – formatting, small edits, quick lookups. Cheap and fast.

on goalso on zenfastcheapopen

Qwen3.7 Max

Alibaba

Open-weights coder tuned for tool use and multi-file edits.

on goalso on zencodingreasoningopen

MiniMax M3

MiniMax

Balanced generalist. Handy fallback that keeps costs predictable.

on goalso on zencodingcheap

MiMo V2.5

Xiaomi

Compact coder for quick edits and autocomplete-style tasks.

on gocodingfastcheapopen

Tip: start with GLM 5.2 as your everyday model. Switch to Kimi K3 for hard problems and DeepSeek V4 Flash for simple, repetitive work. Use /models inside OpenCode to switch whenever the task changes.

Part 01 of 03

Install & subscribe

part 01

01Install & subscribe

Ten minutes to a working setup. No account juggling.

Illustration of a terminal window with colored code
Step 01 / 03
≈ 3 min

Install OpenCode

OpenCode is an open-source coding agent that lives in your terminal - think of it as a calm pair programmer that reads your project and proposes edits. Install it once, then run opencode inside any project folder to start a session.
install opencode
$ curl -fsSL https://opencode.ai/install | bash
Step 02 / 03
≈ 2 min

Subscribe to OpenCode Go

OpenCode Go is a single that unlocks a curated set of good coding models (GLM, Kimi, DeepSeek, Qwen, MiniMax, and more). No juggling API keys from five providers, no local server to run. One price, every model included.
opencode go pricing
# one flat subscription. that's it.
$10  / month              (standard price)
 $5  / first month        (intro pricing for new members)
Step 03 / 03
≈ 1 min

Connect OpenCode to Go

Once you have a Go account, sign in from inside OpenCode. That single command hooks up every included model – no config file, no environment variables.
sign in to go
$ opencode auth login
Part 02 of 03

Use it safely

part 02

02Use it safely

Pick a model, commit your work, run your first session.

Illustration of a small robot sorting colorful cards into trays
Step 01 / 05
≈ 2 min

Choose your coding model

Inside OpenCode, type /models to see every model your Go plan includes. A friendly starting mix:
  • GLM 5.2 or Kimi K2.7 Code – everyday coding
  • Kimi K3 – complex reasoning, architecture, tricky bugs
  • DeepSeek V4 Flash – cheap, simple, repetitive work
inside opencode
$ /models
Step 02 / 05
≈ 2 min

Prepare your repository

Before you let an agent touch your code, make sure you can undo. Commit anything you care about, then run /init inside OpenCode so it can index the project. This is the single most important safety habit in the whole MasterClass.
safety checkpoint
$ git add -A && git commit -m "checkpoint before opencode session"
opencode
/init
Step 03 / 05
≈ 5 min

Run a safe first session

Start with something reversible. Ask the agent to explain a file, add a comment, or write a small test. Review every diff before you accept it – OpenCode always shows changes first.
your first prompt
$ opencode
> explain what src/utils.ts does, then add jsdoc comments
Step 04 / 05
≈ 1 min

Switch models during your session

You are always in charge. When a task changes shape – say you move from scaffolding to a thorny bug – run /models again and pick Kimi K3. Nothing routes automatically; you choose, on purpose.
Step 05 / 05
≈ 2 min

Understand usage limits

Your $10 / month subscription includes rolling usage limits, expressed in dollars of retail inference value (roughly $12 per 5-hour window, $30 per week, $60 per month). Those dollar figures are not extra charges – they are how much model usage is included before Go asks you to wait for the window to reset. Your bill stays $10.
** try it – pick a model for the task (all on opencode go)
Recommended model
GLM 5.2

Fast and cheap. Ideal for scaffolding, tests, and repetitive edits.

How to switch inside OpenCode
/models

Type this in OpenCode, then pick GLM 5.2 from the live list. The exact label on Go may vary slightly.

Illustrative only – Go doesn't auto-route. You stay in charge of which model is answering.

The whole subscription
$10/ month
$5 first month

That is the complete price. The dollar figures in the OpenCode console are the retail value of inference included in the subscription, not extra charges. No separate $60 bill.

Why you see "$60"

For most models, Go targets up to ~$60 of direct-model usage per month for the $10 subscription (~6× value), funded by volume discounts and reserved GPU capacity passed to Go subscribers.

Why counts differ per model

Each model has a different operating cost, so the fixed allowance stretches further on efficient models like DeepSeek V4 Flash. Kimi K3 includes ~$15 instead of $60, still more usage value than buying directly.

When Zen actually charges

Only if you separately fund a Zen balance and enable Use balance. With it off, Go stops at the limit rather than billing more.

How Go compares
PlanMonthlyvs Go
OpenCode Go$10 ($5 first month)–
Claude Code / Codex / Antigravity · entry$202×
Claude Code / Codex / Antigravity · serious dev$10010×
Claude Code / Codex / Antigravity · top tier$20020×

Not identical – frontier plans bundle chat, research, and media. For coding-first work, Go wins on price.

Watch the metered overages

Claude Code also charges per-token usage on top of its subscription. Those metered fees can push the true monthly cost 100× or more above the listed plan pricefor heavy coding sessions. The $20-$200 tiers are starting points, not hard caps.

A practical model portfolio
Efficient models

GLM 5.2, Kimi K2.7 Code, DeepSeek V4 Flash, Qwen, MiniMax – most everyday coding.

Kimi K3

Architecture, hard debugging, long-horizon reasoning.

OpenCode Zen

Only when GPT, Claude, or Gemini is genuinely needed.

Part 03 of 03

Extend & protect

part 03

03Extend & protect

Zen for when Go isn't enough – and the safety habits that make everything reversible.

Illustration of routing paths between two glowing nodes
Step 01 / 02
≈ 3 min

Add OpenCode Zen only when you need it

OpenCode Zen is an optional companion for models Go doesn't include - Gemini, Grok, extra open-weights variants – billed pay-as-you-go from a prepaid balance. You do not need it to start; add it only when a specific task keeps pushing you past the Go catalog.
Step 02 / 02
≈ 3 min

Protect your code and data

A few non-negotiables from the MasterClass:
  • Commit before every substantial agent session. If a diff goes wrong, use git restore or git reset --hard HEAD.
  • Never paste secrets, keys, or client data into a free-tier model. Treat prompts like outgoing email.
  • Read every diff. The agent is fast; you are the reviewer.
safety escape hatch
$ # something went sideways?
git restore .                    # discard unstaged edits
git reset --hard HEAD            # nuke everything back to last commit
** stuck? – a few things that trip people up
recommended crafted intelligence setup

The MasterClass 12 stack

Primary service
OpenCode Go
Everyday model
GLM 5.2 or Kimi K2.7 Code
Complex reasoning
Kimi K3
Simple & repetitive
DeepSeek V4 Flash
Optional expansion
OpenCode Zen
Required safety
Commit before every substantial agent session

This gives members an affordable, flexible coding environment – no premium Claude or OpenAI subscription, no locally maintained routing server, and no separate account for every model provider.

Advanced · bonus chapter

One workspace, many models

advanced · switching between go and zen

One session. many model choices.

Go and Zen are two separate providers inside the same OpenCode app. Connect both once, then move between their models with a single command – no closing the terminal, no copying code, no switching apps.

Connect both services

Run /connect inside OpenCode, pick OpenCode Go, then run it again and pick OpenCode Zen. Both use your OpenCode credentials but appear as separate providers.

connect providers
$ /connect      # pick: OpenCode Go
/connect      # pick: OpenCode Zen
list every model
$ /models
# Go models   → opencode-go/<model-id>
# Zen models  → opencode-zen/<model-id>
what stays the same
  • Your repository & files
  • Git status & working tree
  • OpenCode tools & permissions
  • Project-level instructions
  • The current conversation
  • Your terminal session
what actually changes
  • Which model answers the next turn
  • Billing source (Go allowance vs Zen balance)
  • Context-window size
  • Tool-use reliability & style
  • Speed and cost per request

The practical model ladder

Tap a rung to see which model to pick and why. Real work rarely stays on one model – climb up for hard thinking, drop back down for execution.

rung 01 · Investigate

A hard-to-reproduce bug. Ask a strong reasoning model to trace it before touching code.

Suggested model → Kimi K3 · via Go

Switching isn't orchestration

The new model inherits the conversation, but it may interpret it differently. Models vary in:

  • Context-window size
  • Tool-use reliability
  • Coding style
  • Attention to prior decisions
  • Error recovery
  • Long-horizon planning

Two prompts make handoffs safe. Ask the outgoing model to summarize, then ask the incoming model to review before writing anything.

before you switch
Before I switch models, summarize the
objective, decisions made, files changed,
unresolved problems, and the next
recommended step.
after you switch
Review the work completed so far, the
current Git diff, and the previous model's
plan. Do not make changes yet. Identify
errors, unresolved assumptions, or missing
tests.
billing · read this before enabling "use balance"

Go stays flat. Selecting a Go model uses the allowance included in your $10 / month subscription – no per-request charge.

Zen is prepaid. Selecting a Zen model bills that request against your Zen balance, priced per model.

The OpenCode console has a Use balance toggle. When it's on, work continues against your Zen balance after a Go allowance is reached. Enable it on purpose – it can create pay-as-you-go charges.

Same agent. Same repo. Same session. Different intelligence.

Wrap up

You're done

** end of MasterClass 12

That's the whole class.

Come back to any step later – your progress is saved on this device. Once every box is checked, you'll have a coding setup that belongs to you.

0 / 10 steps complete
organized by Sam Alemayehu · special guest Ariya Hidayat