I currently pay for three AI subscriptions: Claude Pro, ChatGPT Plus, and Google AI Pro. That’s $60 a month, and for coding, only one of them is earning its share.
If you’re choosing an AI subscription for coding, all three sit at about $20 and all three now include a coding agent, a tool that reads your project, edits files, and runs commands on its own. On paper they look interchangeable. In daily use they really aren’t.
Here’s what each plan actually gives you, where each one let me down, and why Claude Pro is the one I’d keep.
What $20 Buys You in Each AI Subscription for Coding
All three bundle a chatbot with a coding agent and share one usage allowance between them, at least for Claude now and ChatGPT soon.
| Claude Pro | ChatGPT Plus | Google AI Pro | |
|---|---|---|---|
| Monthly price (US) | $20 ($17 billed yearly) | $20 | $19.99 |
| Coding agent | Claude Code | Codex | Antigravity, plus the Jules cloud agent |
| Top model for coding | Opus 5.5 | GPT-6 Astra | Gemini 4 Argon, on a separate weekly quota |
| How limits work | Five-hour window plus a weekly cap, shared with chat | Five-hour window plus weekly limits, measured in tokens | Five-hour window plus weekly caps, separate Argon quota |
| Image generation | None | Yes | Yes |
| Extras | Projects, Cowork | Deep research, projects | 5 TB storage, Gemini across Google apps |
Prices are the US monthly prices as of November 2026. Limits are the part that changes most often, so check before you subscribe.
Claude Pro: The One I’d Keep
Anthropic’s pricing page lists Claude Code inside the $20 Pro plan. There’s no separate coding subscription to buy. Chat and Claude Code draw from the same pool, so a long chat session in the afternoon leaves less room for coding in the evening.
That pool used to be my biggest complaint. I hit the limit in Claude Code all the time.
Since Opus 5.5, that has mostly stopped. When Anthropic launched Opus 5.5 on September 22, it said it was raising the five-hour usage limits on Pro, Max, and Team plans. Anthropic didn’t say by how much, but I can feel the difference.
I let AI write most of my code, so the agent matters more to me than the chatbot. Claude Code is the one I trust to change a real codebase without me checking every line twice. I also use Claude through my work account, so this isn’t a one-off impression.
The weak spot is anything visual. Claude doesn’t generate images at all. If you need a quick illustration for a README or a mockup, you’ll be using something else for that.
ChatGPT Plus: Codex Is Good, the Meter Isn’t
I recently subscribed to ChatGPT Plus to try Codex, OpenAI’s coding agent. On Plus it runs in the terminal, in an editor extension, on the web, and on iOS.
The code is fine. The meter is the problem.
OpenAI publishes rough estimates of how many local Codex messages fit in one five-hour window on Plus. In its help article on GPT-6 Astra usage, the estimate for Astra, its top model, is 5 to 45 messages. GPT-6.1 Sol, the default, gets 15 to 160. Those are wide ranges, and OpenAI is clear they aren’t fixed limits.
That matches what happens to me. Whenever I pick the top-tier models in Codex, I end up using all my available tokens.
I’ve also noticed Codex using more tokens than Claude Code for the same coding tasks. That’s my observation from my own projects, not a controlled benchmark, so take it as one developer’s experience.
Two changes are worth knowing about:
- The five-hour limit is back. OpenAI removed it for Plus over the summer, then restored it on August 25, so Plus now has both a five-hour window and weekly limits.
- Chat and Codex are merging. OpenAI’s usage limits article says that starting in December, Plus usage will be shared across ChatGPT and Codex and measured in tokens. Claude already works this way.
Where ChatGPT wins is images. For generating pictures, ChatGPT is the best of the three in my experience. If your work mixes code with design assets or blog graphics, that alone can justify Plus.
Google AI Pro: A Great Bundle With a Shaky Coding Deal
Google AI Pro is the best value of the three if you count everything in the box: 5 TB of storage, Gemini inside Gmail and Docs, image generation, and higher limits in Antigravity, Google’s agent-first coding platform.
I pay for it and I use Antigravity for dev work. But the coding side has been unstable this year.
- Gemini CLI is gone for individuals. On June 18, Google moved AI Pro, Ultra, and free accounts to the new Antigravity CLI. Enterprise licenses and API keys kept Gemini CLI.
- Gemini 3.1 Pro disappeared. Around October 12, it was retired from consumer plans and dropped out of Antigravity’s model picker with little warning. Google’s developer forum filled up with angry AI Pro subscribers.
I lost the Pro-class Gemini models in Antigravity in October. I still have them in the Gemini app.
Google did respond. It rolled Gemini 4 Argon into Antigravity for AI Pro and Ultra subscribers, naming the desktop app and the CLI, with a separate weekly-only quota. Better than nothing, but a weekly-only cap on your strongest model means you ration it.
So why is Google still in this race at all? In my opinion, because Gemini comes preinstalled on Android phones. I covered how much that default matters in my AI assistant for Android post. Plenty of people already pay for Google AI Pro for storage and phone features, and Antigravity comes along for free. That’s a fine reason to have it. It’s a weak reason to pick it for coding.
Can You Skip the Subscription and Code Locally?
It’s the obvious question, and I tried it.
I ran Qwen Coder locally through Ollama, both in plain chat and through the Continue extension in VS Code. My laptop didn’t have the resources. Writing a Go hello world took around 30 minutes. I’ve since uninstalled Ollama.
Local models do make sense on stronger hardware, especially for private code and autocomplete. I wrote about where they fit in my local AI models post. On an ordinary laptop, though, they won’t replace a $20 coding agent.
Pro Tip: Whichever plan you pick, check your usage before you start a big task, not after you hit the wall. Claude Code has
/usage, and Codex and Antigravity both have usage pages. A big refactor started at 90% of your five-hour window is how you lose your context halfway through.
Picking Your $20 AI Subscription for Coding
- You mostly write code with an agent: Claude Pro. Best results for me, and since Opus 5.5 the limits have stopped getting in the way.
- You need images as much as code: ChatGPT Plus. Stick to GPT-6.1 Sol for everyday work and save Astra for the hard problems.
- You live in Google’s apps and code on the side: Google AI Pro. The storage and Android integration carry it, and Antigravity is a decent extra.
- You do it all and can afford two: Claude Pro for coding plus one of the others for images. That’s roughly where I’m heading.
How you work with these tools also changes how fast you burn through limits. My CLI vs GUI AI coding post covers habits that cut token use, like clearing sessions between unrelated tasks.
Where My Money Goes
Three subscriptions is too many, and writing this made that obvious.
Claude Pro stays, because it’s the one I code with. Google AI Pro stays for now, mostly for things that have nothing to do with coding. ChatGPT Plus has to prove that its images are worth $20 a month, because Codex alone hasn’t.
Frequently Asked Questions
Is Claude Code included in Claude Pro?
Yes. Claude Code comes with the $20 Pro plan and shares one usage pool with regular Claude chats. There’s no separate Claude Code subscription.
Is Codex included in ChatGPT Plus?
Yes. Plus includes Codex in the CLI, the IDE extension, the web, and iOS. How much you get per five hours depends heavily on which model you pick.
Why do I hit Codex limits so fast?
The top model, GPT-6 Astra, uses your allowance much faster than the others. OpenAI estimates 5 to 45 local messages per five hours on Plus with Astra, versus 15 to 160 with GPT-6.1 Sol.
Does Google AI Pro still include Gemini CLI?
No, not for individual accounts. Google replaced it with the Antigravity CLI on June 18, 2026. Gemini CLI still works with Enterprise licenses and paid API keys.
Which AI subscription is best for coding on a budget?
Pick one, not three. If coding is your main use, Claude Pro gives the most dependable agent for $20. If you already pay for Google storage, AI Pro adds Antigravity at no extra cost.
