OpenCode in plain wordsNo tech talk

Your Mac can think for free.

Think of OpenCode like a workshop. Some helpers live in your Mac and cost nothing. Some helpers live far away and send a small bill. Start with the free ones.

On your MacFree and private
Free far awayAlso $0
Paid far awayOnly when it matters
Hokusai wave, part of the Milo Offline visual system
The simple rule: use the free helper first. If the job is too hard for them, call in the paid expert.

On this page: How to use · Every helper · By task · Devotional example · Fallbacks · Families · Sources

How to use the free one on your Mac.

Do this once. It takes five minutes. After that, you just talk to it.

  1. Open the black box.On your Mac, open the app called Terminal. That is the black box where you type orders.
  2. Start OpenCode.Type the word opencode and push Enter. Wait for it to say hello.
  3. Pick your helper.Type /models and push Enter. Pick the one called ollama / devstral-local-safe. That is the free one that lives in your Mac.
  4. Just talk.Type what you want in plain words. Like: tidy up my draft. It answers right there.
  5. Change helpers any time.Type /models again and pick a different name from this page. Nothing breaks. You can always go back to the free one.
  6. Leave when done.Type /exit or just close the black box. Your Mac helper sleeps until you need it.

Every helper, plain and simple.

All 78 names OpenCode offers today. Cheapest first. Plain words for each group.

0

Lives in your Mac. Free forever.

Stays home. Sees nothing but your Mac. Good for daily writing, fixing, and private drafts.

  • ollama / devstral-local-safe — your everyday Mac helper. The default. Good at fixing files and writing drafts.
1

Free far away. Costs $0.

Lives far away but the maker lets you use them free. Good for short chores. A little slower than the Mac one.

  • opencode / ling-3.0-flash-fin-free — quick small chores.
  • opencode / longcat-2.5-preview-free — quick drafts and rewrites.
  • opencode / mimo-v2.6-flash-free — neat summaries and sorting.
  • opencode / nemotron-3-ultra-free — stronger free thinking.
  • opencode / nemotron-3.5-lightning-free — fast free answers.
  • opencode / space-bunny-free — light fun jobs.
  • opencode / muse-spark-1.3-contributor-free — general chat, free. This is what answers you right now.
2

Penny helpers. Almost free.

Each job costs a few cents or less. Covered by plans you already pay for, or too small to notice.

  • opencode / big-pickle — playful small model, tiny jobs.
  • opencode / deepseek-v4-flash — fast cheap writing.
  • opencode / deepseek-v4-flash-vision-exp — same, can also look at pictures.
  • opencode / deepseek-v4.1-flash — newer fast cheap writing.
  • opencode / gemini-3.5-flash-lite — Google's lightest, cheapest.
  • opencode / glm-5.3-flash — fast list-making and forms.
  • opencode / gpt-5-nano — tiniest ChatGPT, labels and sorting.
  • opencode / gpt-5.4-mini — small ChatGPT, short summaries.
  • opencode / gpt-5.4-nano — tiniest new ChatGPT.
  • opencode / gpt-5.1-codex-mini — tiny coding helper.
  • opencode / gpt-5.3-codex-spark — quick code ideas.
  • opencode / gpt-5.6-luna — light ChatGPT, cheap and quick.
  • opencode / gpt-6-luna — newest light ChatGPT.
  • opencode / gpt-5.6-sol — your proven devotional checker. Free inside your ChatGPT plan.
  • opencode / grok-build-0.1 — early test model, tiny jobs only.
  • opencode / muse-spark-1.2 — general chat, paid pennies.
  • opencode / muse-spark-1.3 — general chat, paid pennies.
  • opencode / qwen3.8-flash — fast Qwen for lists and drafts.
3

Everyday workers. Small bill.

Strong enough for real work. A few dollars a month if used daily. This is where most jobs should land when free is not enough.

  • opencode / claude-haiku-4-5 — fastest Claude, cheap reviews.
  • opencode / claude-sonnet-4 — balanced Claude.
  • opencode / claude-sonnet-4-5 — balanced Claude, newer.
  • opencode / claude-sonnet-4-6 — balanced Claude, newest of this line.
  • opencode / claude-sonnet-5 — strong all-rounder.
  • opencode / gpt-5 — full ChatGPT.
  • opencode / gpt-5-codex — full ChatGPT for code.
  • opencode / gpt-5.1 — newer ChatGPT.
  • opencode / gpt-5.1-codex — newer ChatGPT for code.
  • opencode / gpt-5.1-codex-max — biggest of that code line.
  • opencode / gpt-5.2 — newer ChatGPT.
  • opencode / gpt-5.2-codex — newer ChatGPT for code.
  • opencode / gpt-5.3-codex — newest code ChatGPT.
  • opencode / gpt-5.4 — current ChatGPT.
  • opencode / gpt-5.5 — current ChatGPT, stronger.
  • opencode / gpt-5.6-terra — strong ChatGPT worker.
  • opencode / gemini-3-flash — fast Google.
  • opencode / gemini-3.5-flash — fast Google, newer.
  • opencode / gemini-3.6-flash — fast Google, newer still.
  • opencode / gemini-3.7-flash — fast Google, current line.
  • opencode / gemini-3.8-flash — fast Google, latest line.
  • opencode / deepseek-v4-pro — full DeepSeek, stronger thinking.
  • opencode / glm-5 — full GLM worker.
  • opencode / glm-5.1 — GLM, newer.
  • opencode / glm-5.2 — GLM, newer still.
  • opencode / glm-5.3 — GLM, current.
  • opencode / kimi-k2.5 — Kimi writer.
  • opencode / kimi-k2.6 — Kimi writer, newer.
  • opencode / kimi-k2.7-code — Kimi for code.
  • opencode / kimi-k3 — newest Kimi.
  • opencode / minimax-m2.5 — MiniMax worker.
  • opencode / minimax-m2.7 — MiniMax, newer.
  • opencode / minimax-m3 — newest MiniMax.
  • opencode / grok-4.5 — Grok worker.
  • opencode / grok-4.6 — Grok, newer.
  • opencode / grok-4.7 — newest Grok.
  • opencode / qwen3.5-plus — full Qwen writer.
  • opencode / qwen3.6-plus — full Qwen, newer.
  • opencode / qwen3.8-max — biggest fast Qwen.
4

Top experts. Real money.

Save these for big days: launch day, money pages, hard problems. Dollars per job.

  • opencode / claude-opus-4-5 — top Claude.
  • opencode / claude-opus-4-6 — top Claude, newer.
  • opencode / claude-opus-4-7 — top Claude, newer still.
  • opencode / claude-opus-4-8 — top of that line.
  • opencode / claude-opus-5 — current top Claude.
  • opencode / claude-opus-5-5 — very top Claude.
  • opencode / claude-fable-5 — biggest storyteller Claude.
  • opencode / claude-fable-5-1 — biggest storyteller, newest.
  • opencode / gpt-5.4-pro — biggest ChatGPT of that line.
  • opencode / gpt-5.5-pro — biggest current ChatGPT.
  • opencode / gpt-6-sol — newest top ChatGPT.
  • opencode / gpt-6-astra — newest top ChatGPT, second flavor.
  • opencode / gemini-3.1-pro — biggest Google, huge memory.

What to use for your real jobs.

Match the job to the helper. Easy jobs stay free. Hard jobs call the expert.

Fixing apps

Mac helper writes the fix. A worker checks it. The expert only comes for scary bugs.

START FREE, CALL EXPERT FOR SCARY BUGS

Writing and devotionals

Mac helper writes the draft. The free far-away checker reads it. The expert gives the final yes.

FREE DRAFT, EXPERT FINAL YES

Brand and design

Mac helper tidies words. A worker makes options. The expert picks the taste.

EXPERT PICKS THE TASTE

Research

Mac helper reads what you hand it. The far-away helpers check what is new today.

MAC READS, FAR-AWAY VERIFIES

Pictures, video, voice

Free Kaggle computers do the heavy lifting. The Mac helper cuts up scripts.

KAGGLE FIRST

Money and launch days

Always let the top expert read anything about money, passwords, or launch day.

ALWAYS CALL THE EXPERT

One real example: your devotional checker.

Your devo-go job has 6 steps. Steps 1 to 4 are automatic and always free. Step 5 is one helper reading the draft. Here is what each helper costs for that one read.

  1. Steps 1 to 4: automatic checks.Your Mac does them in a blink. Always $0. No helper needed.
  2. Step 5: one helper reads the draft.About a book chapter of reading. Takes about a minute.
  3. Step 6: the gate.Automatic again. It stops the job if the reader found big problems.

Last real run: 67 seconds, found 6 problems, 2 big ones, paid $0.00. Full record in your 2026-09-25-real-model files.

When your favorites run out.

Flagships and plan models have usage windows. When one says slow down or stop, move down this list. Do not stop working. Just step down one rung.

Read it like this

Each job below names its first pick, then what to use when the first pick is tired. The last name on each row still gets the job done. It is just slower or less sharp. Anything marked never means it will waste your time on that job.

1

Fixing code

Offline OS, apps, tracker, game fixes.

  • First: ollama / devstral-local-safe — free, private.
  • If Mac is busy: opencode / gpt-5.1-codex-mini or opencode / claude-haiku-4-5.
  • If those are out: opencode / kimi-k2.7-code, then opencode / qwen3.8-flash.
  • Top expert only for scary bugs: opencode / claude-opus-5, then opencode / gpt-6-sol.
  • Never: picture-only test models, storyteller flagships for a one-line fix.
2

Checking work (devo-go step 5, audits)

Needs a strict reader. This is the one job where weak readers cost you.

  • First: opencode / gpt-5.6-sol — free in your plan, scored 1.00 last time.
  • If the plan is tired: opencode / claude-sonnet-5, then opencode / gpt-5.5.
  • If money matters: opencode / mimo-v2.6-flash-free catches most, misses nuance.
  • Big days only: opencode / claude-opus-5-5, then opencode / claude-fable-5-1.
  • Never: the Mac helper alone for a final launch devotional.
3

Writing drafts (devotionals, briefs, docs)

Drafts are cheap. Final yes is not.

  • First: ollama / devstral-local-safe or opencode / muse-spark-1.3-contributor-free.
  • If you want smoother: opencode / kimi-k3, opencode / qwen3.6-plus, opencode / minimax-m3.
  • If those are out: opencode / gemini-3.6-flash, then opencode / gpt-5.4-mini.
  • Final yes: opencode / claude-fable-5 or opencode / gpt-5.5-pro.
4

Brand taste and final words

Taste is flagship work. Do not let a small model publish alone.

  • First: opencode / claude-opus-5 or opencode / claude-fable-5-1.
  • If flagships are tired: opencode / claude-sonnet-5, then opencode / gpt-5.5, then opencode / grok-4.7.
  • Workers prep, never publish: opencode / minimax-m2.7, opencode / glm-5.3.
  • Never: nano, lite, or flash models for the final brand call.
5

Research and fact checks

Small models guess. Only big connected ones verify.

  • First: opencode / gemini-3.1-pro or opencode / gpt-6-astra.
  • If tired: opencode / claude-sonnet-5, then opencode / gpt-5.4, then opencode / kimi-k3.
  • Mac helper only reads what you hand it. It cannot check what is new today.
  • Never: free tiny models for laws, prices, or live product facts.
6

Long piles (big repos, long docs)

Needs a big memory. Small models forget the start.

  • First: opencode / gemini-3.1-pro, then opencode / qwen3.6-plus, then opencode / kimi-k3.
  • If tired: opencode / gemini-3.8-flash, then opencode / minimax-m3.
  • Mac helper caps at 64K. Split the pile before handing it to them.
7

Pictures and screenshots

Only some helpers can see.

  • First: opencode / gemini-3.6-flash or opencode / deepseek-v4-flash-vision-exp.
  • If out: opencode / qwen3.8-flash, then opencode / gemini-3.8-flash.
  • Never: plain text models. They are blind.
8

Quick questions all day

Protect the flagships. Spend free answers here.

  • First: ollama / devstral-local-safe.
  • Then free far away: opencode / muse-spark-1.3-contributor-free, opencode / nemotron-3.5-lightning-free, opencode / longcat-2.5-preview-free.
  • Then pennies: opencode / gpt-5-nano, opencode / gemini-3.5-flash-lite.
  • Never call a flagship for what time is it.

Every family, and when to call them.

One card per maker. What they are like, when they shine, and when to stop and call someone bigger.

Ollama (your Mac)

HOME TEAM

One helper: devstral-local-safe. Private coder. Writes fixes, tests, drafts. Costs nothing.

Call when
Daily work, private material, drafts you will review.
Stop when
Final judgment, live facts, launch-day taste.

Claude (Anthropic)

BEST JUDGMENT

Haiku is fast and cheap. Sonnet 4 through 5 are balanced workers. Opus 4-5 through 5-5 and Fable 5 through 5-1 are the top experts.

Call when
Hard bugs, architecture, final copy, final review.
If tired
Step Opus to Sonnet 5 to Haiku. Never step a final down to free.

GPT / Codex (OpenAI)

PROVEN REVIEWER

Nano and mini for tiny chores. 5 through 5.6-terra for everyday work. Codex lines for code. 5.6-sol is your plan-covered devo checker. 6-sol, 6-astra, and the pro models are flagships.

Call when
Code review, strict checklists, devo step 5, hard reasoning.
If tired
Step pro to 5.5 to 5.4-mini to nano. Keep sol for daily checks.

Gemini (Google)

BIG MEMORY

Flash-lite and flash lines for fast cheap work. 3.1-pro is the giant memory for huge piles and research.

Call when
Long docs, big repos, screenshots, current research.
If tired
Step 3.1-pro down through 3.8, 3.7, 3.6 flash.

DeepSeek

CHEAP THINKER

Flash lines for pennies. Vision-exp can also see pictures. Pro for deeper thinking on a budget.

Call when
Analysis and drafts when money is tight.
If tired
Step pro to 4.1-flash to 4-flash.

Qwen (Alibaba)

FAST ALL-ROUNDER

3.8-flash for speed. 3.5-plus and 3.6-plus for full writing. 3.8-max for the biggest fast jobs.

Call when
Drafts, rewrites, lists, second opinions.
If tired
Step max to plus to flash to the Mac helper.

GLM

STEADY WORKER

5 through 5.3 for everyday work. 5.3-flash for the cheap fast lane.

Call when
Forms, lists, structured jobs, mid-size writing.

Kimi

WRITER

2.5 and 2.6 for writing. 2.7-code for code. K3 is the newest and strongest.

Call when
Long reads, clean writing, code help on a budget.

MiniMax

BATCH HAND

M2.5 through M3. Good hands for piles of similar jobs.

Call when
Variants, cutdowns, archives, repeatable batches.

Grok (xAI)

BOLD VOICE

4.5 through 4.7 for strong drafts with edge. Build-0.1 is an early test, tiny jobs only.

Call when
Fresh angles, bold drafts, second opinions.

Muse Spark

HOUSE CHAT

1.2 and 1.3 for penny chat. The contributor-free one answers free.

Call when
Everyday questions when the Mac helper is busy.

Nemotron (Nvidia)

FREE MUSCLE

Ultra-free and lightning-free. Surprisingly strong for $0.

Call when
Free second opinions and summaries.

Free small teams

FREE BENCH

Ling fin-free, Longcat preview-free, Mimo flash-free, Space-bunny-free. Light, fast, narrow.

Call when
Sorting, tagging, short summaries, routing.
Stop when
Anything final, theological, legal, or money-related.

Big Pickle

WILD CARD

Playful small model. Fun tiny jobs. Never load-bearing work.

Call when
Play, experiments, throwaway drafts.

Where this comes from.

Model names copied from OpenCode on this Mac today. Prices from your plan papers. Devo numbers from your real September 25 run.

Back to the main ecosystem pageHow the Mac helper plugs into OpenCodeAbout the Mac helperMilo Offline brand system

Free offers and prices change often. When in doubt, start free and move up only if the answer is not good enough.