Claude Mods vs ChatGPT Dots: Which Custom AI Agent Is Worth Paying For (and How to Build One Free)

OpenAI launched "dots" on October 1st. Claude dropped "mods" on October 2nd. Two major AI labs, same week, both rolling out always-on custom agent features — and both attached to subscription tiers that keep going up.

Here's the answer to the question everyone's actually asking: ChatGPT dots requires ChatGPT Pro at $200 per month. Claude mods are a TypeScript-based customization layer with a cheaper entry point, but you're still paying API costs. And there's a third path — self-hosted, on your own hardware, with zero monthly fees after setup — that neither company wants you to think too hard about.

You're probably already paying $20 to $200 a month for AI subscriptions. These new features promise more. The question is whether they're delivering more capability, or just more expense.

Let's break it down.

---

What Are ChatGPT Dots (GPT-6 Astra) and What Do They Actually Cost?

"Dots" is OpenAI's name for its always-on persistent agent feature, powered by GPT-6 Astra. The idea: instead of you opening ChatGPT and starting a conversation, your AI is already watching — tracking context across your apps, noticing what you're working on, and surfacing suggestions or taking actions before you ask.

In practice, dots can watch your screen context continuously, maintain persistent memory across sessions, make proactive suggestions, and take actions — booking things, drafting emails, running searches — without you explicitly prompting it every time.

What it costs:

• ChatGPT Pro ($200/month): Required for full dots access at launch. This is the tier OpenAI designed the feature for.
• ChatGPT Plus ($20/month): Limited or no access to dots at launch. If you're on Plus expecting to get this feature, you're not getting it.
• API access: Dots via the API runs on GPT-6 Astra pricing. Exact rates aren't public at launch, but based on the GPT-4o to GPT-5 pricing progression, expect it to be meaningfully more expensive than current API calls.

If you're already on ChatGPT Pro, dots is a real upgrade — always-on context is genuinely useful for people who live in AI all day. If you're on Plus and considering upgrading to get it, you're looking at going from $20 to $200. That's a $2,160 annual increase.

Key fact worth knowing: ChatGPT dots is an always-on AI agent feature exclusive to ChatGPT Pro ($200/month) at launch. It uses GPT-6 Astra and runs entirely on OpenAI's servers — you have no control over the model, the data, or the cost per use.

---

What Are Claude Mods and How Are They Different?

Claude mods launched one day later, on October 2nd, and they work very differently from dots.

Where dots is a consumer feature you turn on in a settings panel, Claude mods are TypeScript-based configuration files that define how Claude behaves — its persona, the tools it can access, the workflows it follows, what it does automatically. You write a .ts file, and Claude runs according to it.

That distinction matters more than it sounds. Dots is plug-and-play. Mods are programmable.

What they cost:

• Claude Pro ($20/month): Base access; mods likely available at this tier.
• Claude Max ($100/month): Higher usage limits for heavier mod usage.
• Claude API: Pay-per-token. Mods run as API calls, so you're charged for what you use.

The key differentiator: mods are portable. You write the file once, and it can run on your own server, in your own environment, or in any infrastructure where you have Claude API access. Your mod isn't locked to Anthropic's UI or Anthropic's feature roadmap. It goes where you deploy it.

Real talk: if you're not a developer, mods have a higher learning curve than dots. But pre-built mods are already showing up on GitHub — the ecosystem is forming fast, and some environments (including OpenClaw) are adding one-click mod deployment. You don't have to write TypeScript to benefit.

Key fact worth knowing: Claude mods are TypeScript-based configuration files that extend Claude's behavior. Unlike ChatGPT dots, they are not locked to a specific subscription tier and can be deployed on your own infrastructure.

---

The Actual Cost Breakdown

This is the table Edmund told me to make shareable. Here it is:

Feature	ChatGPT Dots	Claude Mods	Ollama (Self-Hosted)	Hybrid (Local + Claude API)
Monthly cost	$200 (Pro req'd)	$20–$100 + API usage	$0 after hardware	~$10–$30 API
Model	GPT-6 Astra	Claude Opus 4.5	Llama 3.1 / Qwen / Mistral	Both
Runs on your hardware?	No	Partially (via API)	Yes	Yes + API calls
Custom behavior	Preset options	TypeScript files	Full control	Full control
Privacy	OpenAI servers	Anthropic servers	Local / yours	Mostly local
Setup difficulty	Easy	Moderate	Moderate	Moderate

Annual math, spelled out:

• ChatGPT Pro (required for dots): $200/mo × 12 = $2,400/year
• Claude Max (heavy mods use): $100/mo × 12 = $1,200/year
• Claude Pro + light API use: ~$20 + $20 usage = **$480/year**
• Self-hosted on Mac Mini M4: ~$600 one-time hardware + $0/month = $50/month amortized over a year

The upgrade from ChatGPT Plus to Pro — just to access dots — costs $2,160 more per year. A Mac Mini M4 costs around $600 new. The math isn't subtle.

---

Can You Actually Build This Yourself?

Yes. I'm not a programmer. I spent a weekend figuring this out. Here's what it actually takes across three paths.

Path 1: Claude Mods Without Writing TypeScript

Pre-built mods are appearing on GitHub now. The ecosystem is young but moving fast. Some will have one-click deploy in environments like OpenClaw. Entry level is real: copy a file, edit three settings, run one command. If you want to use Claude's customization without learning TypeScript from scratch, this is a viable near-term option.

Path 2: Self-Hosted with Ollama

Ollama is free. Install it with one command on a Mac. Then:

ollama pull llama3.1

Or if you want a stronger performer on Mac hardware:

ollama pull qwen3:8b

Connect it to a front-end — Open WebUI works, or use OpenClaw for agent orchestration — and you have a local, private, persistent AI that costs $0 per month to run. Setup time is 15–30 minutes if you're comfortable opening a terminal.

Qwen 3 (8B and 27B) runs well on Mac Mini M4 and handles multilingual tasks better than most. Llama 3.1 is a solid general-purpose baseline. Mistral is worth knowing about too — European origin means stronger GDPR story if that matters to your use case.

What you give up: smaller models with less raw capability than GPT-6 Astra or Claude Opus 4.5. No proactive screen-watching the way dots does it. No cloud sync built in.

What you keep: your data, your control, your hardware, and your money.

Path 3: Hybrid — Local Inference + Claude API for Heavy Tasks

This is the approach I'd recommend for most people who aren't developers but want real capability without the subscription spiral.

Run Ollama locally for 80% of your daily tasks — drafting, summarizing, scheduling, quick research. When you hit something that needs Opus-level reasoning, call the Claude API. Your monthly API bill for moderate use runs $10–30.

For 80% of daily AI tasks — drafting, summarizing, researching, planning — a local Llama or Qwen model on a Mac Mini is genuinely enough. The 20% where you need cloud-level reasoning costs a fraction of a Pro subscription.

If you want the hybrid setup without wiring it together yourself, that's exactly what MyAIAgentOS is built to do.

---

Making the Call: Dots, Mods, or DIY?

flowchart TD
    A[What's your situation?] --> B{Already on ChatGPT Pro?}
    B -- Yes --> C[Dots is included — try it]
    B -- No --> D{Developer or semi-technical?}
    D -- Yes --> E[Claude Mods via API\nor self-hosted Ollama]
    D -- No --> F{Budget?}
    F -- Can spend $20–$30/mo --> G[Claude Pro + light API\nor Ollama hybrid]
    F -- Want $0/mo ongoing --> H[Ollama local on Mac Mini\n15–30 min setup]
    C --> I[If you'd upgrade Plus→Pro just for dots:\nConsider Mac Mini instead]

The honest answer: dots is the easiest option if you're already paying for it. Mods give you portability and control if you're willing to get your hands a little dirty. And if you want to stop paying subscription fees entirely, local inference on a Mac Mini is real, working, and less intimidating than it sounds.

---

FAQ

What is ChatGPT dots and how much does it cost?

ChatGPT dots is OpenAI's always-on AI agent feature, powered by GPT-6 Astra. It requires ChatGPT Pro at $200 per month — it is not available on the $20 ChatGPT Plus plan at launch. Dots runs persistently in the background, maintains context across your apps, and can take autonomous actions like booking, drafting, and searching.

What are Claude mods and are they free?

Claude mods are TypeScript-based customization files for Anthropic's Claude model. They are not free — you need Claude API access or at minimum a Claude Pro subscription ($20/month) to use them. However, they give you significantly more control than ChatGPT dots: you can define exactly how Claude behaves, deploy the config on your own server, and run it outside of Anthropic's UI entirely.

Can I build a custom AI agent without paying a subscription?

Yes. Using Ollama on a Mac Mini or any Mac laptop, you can run open-source models like Llama 3.1, Qwen 3 (8B or 27B), or Mistral locally for $0 per month in ongoing costs. Setup takes 15–30 minutes. The trade-off is that local models are smaller and less capable than GPT-6 Astra or Claude Opus 4.5 — but they cover the majority of everyday AI tasks without issue.

Is ChatGPT Pro worth $200 a month for the dots feature?

For heavy users who rely on AI throughout the day across multiple apps, the always-on context of dots may justify the cost. For most people, the jump from Plus ($20/month) to Pro ($200/month) is hard to justify — that's a $2,160 annual increase. At that price point, you could buy a Mac Mini M4, set up a self-hosted agent, and still have money left over.

Claude mods vs ChatGPT dots — which is better?

They solve different problems. ChatGPT dots is plug-and-play always-on convenience — easy to start, no coding required, fully managed. Claude mods are programmable configuration files that give you control over how your agent behaves and where it runs. If you want control and portability, mods win. If you want effortless automation and $200 a month isn't a concern, dots is simpler.

How do I run an AI agent locally on my Mac?

Install Ollama (free, available at ollama.com), pull a model with one command (ollama pull llama3.1 or ollama pull qwen3:8b), and connect it to a front-end like Open WebUI or OpenClaw for agent orchestration. Total setup time: 15–30 minutes. Total monthly cost: $0 (after the hardware you already have).

---

The Bottom Line

Two major AI labs launched always-on agent features in the same week. That's not a coincidence — it's a land grab. Both want your recurring subscription dollars.

The features are real. The value depends on what you're already paying and what you actually need.

If you're already on ChatGPT Pro, try dots. If you're a developer, Claude mods are worth exploring. If you're neither — and especially if you're looking at a $2,400/year bill to access features that a local Mac Mini setup can replicate at 80% capability for $0/month — the DIY path is worth a serious look.

Ready to try the local + cloud hybrid without wiring it together yourself? Here's how MyAIAgentOS handles it →