When ChatGPT, Claude, Gemini, or Grok is simply easier

Tomás closed the MacBook Air at 11:40pm on Friday with three model files on a 256 GB disk and an 11-line thank-you that still read like a user manual. Notes.app held six bullets about Northline Cold, a vendor who had delivered 14 pallets a day late and then recovered the Friday dock slot. Operations wanted a warm note that kept the time. Tomás had spent Wednesday, Thursday, and Friday evenings in Ollama, swapping llama3.1:8b for qwen2.5:14b (7.3 GB, about 40 seconds per paragraph on 16 GB of RAM) and pasting a system prompt from a Discord screenshot. Monday at 9:40am, Lina in sales dropped the same six bullets into Claude. Ninety seconds later the account manager sent the draft. Tomás still had 19.4 GB of GGUF files and a cafe-wifi pull that had taken 11 minutes on Wednesday.

This is OS7, Part 7 of Open-source AI explained. Previous: OS6, random model downloads. Next: OS7b, open-weight vs open-source vs free. The product door still lives in Which AI product should I use?. This page is the pause after OS2’s location split: a weight file is a tool you load. ChatGPT, Claude, Gemini, and Grok are products. For everyday writing, tools, memory, and an app that works without a fan, a Plus or Pro chat around $20 (as of writing, August 2026) is usually the right buy. Groq (the inference host, spelling with a q) is not Grok (xAI’s chat).

When the hosted chat is enough

• Why a GGUF on disk does not replace a closed chat product
• What Plus, Pro, and SuperGrok actually buy (writing, tools, memory, the app)
• A job table: open or local vs closed, with an honest pick per row
• When local still wins: raw files, huge volume, offline, learning
• A two-minute checklist, and how Tomás spent three evenings on a 90-second job

A weight file is a tool you load

Open weights are a file (or a pile of shards) plus a license. You can run them in Ollama, LM Studio, or llama.cpp, or you can send prompts to a host that already has the GPUs. That is capability: hardware, a runner, a quantization, and a card you should read in later parts.

Closed chats are products. ChatGPT, Claude, Gemini, and Grok sell a box that drafts ordinary English, keeps custom instructions, takes a PDF, sometimes browses, and works on a phone in a parking lot. The model under the box changes. The product job does not. Tomás treated qwen2.5:14b as if the download would give him Claude’s first-draft manners. The file gave him tokens on a fan. Lina used the product.

Rule of thumb: If the text can leave the building and the job is a paragraph, pay the chat plan. Save the local stack for files that cannot leave, for offline, for volume, or for the lesson.

People reach for Ollama after OS1 because “open” felt like more control. Control is real when the prompt stays on the machine. A 14B quantized model can write. On a 16 GB Air it will not match the product Tomás was shadowing, and it will not remember that Operations hates the phrase “per my last email.”

What the closed chat actually buys

As of writing (August 2026), ChatGPT Plus and Claude Pro sit around $20 a month. Google’s everyday paid rung (often Google AI Pro) sits near $19.99. SuperGrok has often sat near $30, a step above the $20 cluster. Names move. Confirm the vendor page the week you click Buy. P7 maps Free vs paid and the three payment rails.

Paid is not a soul transplant. You buy four practical things.

• Everyday writing that is already tuned for email, outlines, and rewrites, without a 7.3 GB pull and a 40-second paragraph.
• Tools the apps already ship: file upload, browsing, code-in-the-sidebar, connectors, projects, custom GPTs. Each vendor’s set moves. You do not wire them.
• Memory and custom instructions, so Monday’s thank-you can sound like last month’s, without a Discord system prompt.
• An app that works. Phone, laptop, a flaky warehouse network. No ollama list, no disk warning at 18 GB free, no cafe-wifi 11-minute pull.

Gemini sits next to Drive and Docs if you already live in Google. Claude Pro, as of writing, is the rung that tends to include extra surfaces (Code, Cowork, and friends: re-check). ChatGPT Plus is the writing box a lot of teams already have. Grok’s paid path lives under SuperGrok and, in some setups, an X subscription. One product for writing is the chooser default. Tomás did not lack a model. He lacked 90 seconds in a box he already could have opened.

Hosted open models (Groq, Together, Fireworks, OpenRouter) are a third door. They rent GPUs for Llama, Qwen, DeepSeek, and friends. The prompt still leaves. That API map is OS7d. If a slide says “we moved to Groq,” ask whether they mean the inference cloud or they misspelled Grok. Wrong vendor, wrong privacy story, two letters.

Job by job

Pick the door by the job, then pick the app. Closed means ChatGPT, Claude, Gemini, or Grok (the product). Open/local means a file you run, or a hosted open-model API if you accept that the prompt leaves.

Learning is a real job. If you want RAM, quantization, and ollama list, do that on a toy prompt. Do not use the Northline note as the toy. Volume is a real job too, and 11 lines is not volume. If you do not have a batch, you have a $20 argument, and the $20 chat already exists.

When local still wins

Local is the right door four times. The internet will try to sell you a fifth (“sovereignty”) that is usually a thank-you with extra steps.

Raw files that cannot leave

Unredacted PDFs, HR, patient-adjacent rows, a customer dump with emails still in column C. P4 is still the paste test: would you email this file to that vendor. OS4 is the open-model version. Local wins only if you stay offline and you do not flip a cloud id in the same window. OS2’s 4.7 GB download did not save Ken once he toggled cloud. Compare quality on a public sentence. Paste work only where the location matches the promise.

Cost at huge volume

A Plus seat is cheap for a person who writes. An API bill is a different animal once you loop a model over 80,000 support tickets. Hosted open APIs can undercut Claude’s API. A machine you already own, running overnight, can undercut both at idle. Do that math with a real count. It does not start with a GPU quote to avoid $20. If you do not have the 80,000, you do not have this row.

Offline, and learning the stack

A plane to Cleveland with no seat-back Wi-Fi is offline. A lab that blocks outbound HTTPS is offline. Then a local file is the only chat you have. Learning is adjacent: you want the runner, the card, the fan, the RAM number from OS3. Label it as homework. Keep the vendor note in Claude. The Run open models from scratch series (OS8 onward) is the how-to once this page has told you the job is actually local.

Switching is allowed. Loyalty to a download is not a KPI.

Worked example: three evenings and 90 seconds

The Northline note was six bullets. One of them still said “sorry about the freezer??” with two question marks. The AM needed 11 lines that kept the Friday slot without sounding like a threat. That is a writing job. It is the chooser’s writing door.

Ollama did what a runner does. The misspent object is three evenings on a job the $20 chat is built for. If Tomás later needs a warehouse CSV that cannot leave, those evenings start to pay. He did not have that CSV on Friday. He had a thank-you.

A two-minute checklist

Five flags. Change them for the job in front of you. This is a toy. It does not call an API. Run it, read the printout, then do that door. Do not add a sixth flag called “but open feels cooler.”

# Two-minute door check. Toy flags only.
# As of writing (August 2026): Plus / Claude Pro ~$20; SuperGrok often ~$30.
answers = {
    "text_must_not_leave": False,  # unredacted files, HR, customer dumps
    "need_offline": False,         # plane, lab, no network
    "learning_the_stack": False,   # the lesson is Ollama / llama.cpp
    "huge_batch": False,           # volume that $20 caps or API $ will feel
    "need_app_tools": True,        # files, browse, memory, a phone app
}

def door(a):
    if a["text_must_not_leave"] or a["need_offline"]:
        return "LOCAL: stay offline. Confirm ollama list has no cloud id."
    if a["learning_the_stack"]:
        return "LOCAL for the lesson. Keep a closed chat for the thank-you."
    if a["huge_batch"]:
        return "LOCAL or a hosted open API. Do the token math this week."
    if a["need_app_tools"]:
        return "CLOSED CHAT: Plus/Pro (confirm price this week)."
    return "CLOSED CHAT. Do not spend three evenings matching a thank-you."

print(door(answers))

What that code prints for Tomás’s thank-you, and for a later CSV that cannot leave:

If you are learning the stack, flip learning_the_stack and still keep the thank-you in Claude. Two jobs, two doors, one evening instead of three.

Open is not a moral upgrade

• Spending three evenings so a local draft matches Claude’s 90-second thank-you.
• Calling Groq (inference) Grok (xAI), or the reverse, then putting the wrong privacy story on a slide.
• Treating a GGUF as a personality, then being angry that memory and file tools are missing.
• Buying a GPU to avoid a $20 chat plan you do not yet have.
• Pasting the unredacted file into Claude “to compare” after you claimed local-only.
• Skipping the closed chat because OS1 made “open” sound like a moral upgrade.

Name the job, then pick closed or open

Take one real job. Fill the five flags. If the printout is CLOSED CHAT, open the one vendor from the chooser and send the note. If it is LOCAL, stay offline, use a toy prompt first, and time one paragraph. Screenshot ollama list so you can see a cloud id if one sneaks in. Next: OS7b. The run-it path starts at OS8 if the job is actually a runner. Learn index: Learn.

Stay on the closed app when it fits

• Load open weights when you need the file. Buy a closed chat when you need the product.
• Everyday writing, tools, memory, and an app that works: Plus or Pro around $20 as of writing (SuperGrok often a rung above). Confirm this week.
• Local still wins for raw files that stay, huge volume, offline, and learning. Groq ≠ Grok.
• Run the five flags in two minutes. Do not spend three evenings matching a thank-you.

Sources

• Ollama (local runner; confirm you are not on a cloud id)
• LM Studio and llama.cpp (desktop window and the engine under many apps)
• Groq (hosted inference, not xAI Grok)
• Together AI and Fireworks AI (hosted open models)
• ChatGPT pricing, Claude pricing, Google One AI plans, and Grok / xAI (plan names and sticker prices as of writing; re-check)
• AMS: Which AI product should I use? and P7 accounts and Free vs paid
• AMS: Open-source AI explained (this series) and OS2 location split

More definitions in Key Terms →

Merch on Etsy

Open store →

You're On Mute White Ceramic Coffee Mug, Data Science Mug, Data Mug, Analytics Mug, Statistics Mug, Programming Mug, Gift Mug, Coffee Mug View on Etsy → Mute Mouse Pad, Data Science Mouse Pad, Analytics Mouse Pad, You're On Mute View on Etsy → Good Cheap Fast White Ceramic Coffee Mug, Data Science Mug, Data Mug, Analytics Mug, Statistics Mug, Programming Mug, Gift Mug, Coffee Mug View on Etsy → Naive Bayes Mouse Pad, Data Science Mouse Pad, Analytics Mouse Pad, Bayesian Mouse Pad View on Etsy →