Every AI Subscription, Ranked: What’s Actually Worth $20/Month in 2026

Every AI company wants $20 of your money every month. ChatGPT, Claude, Gemini, Perplexity, Copilot, Grok. They all landed on the same magic number.

Give $20 to each one and you’ll get six wildly different experiences. Some are worth it. Some are stealing from you. I’ve been deep in this ecosystem for years, and I’m going to break down exactly what you get at each price point so you can stop guessing.

The $20 Landscape

Here’s the wild part. Six major AI companies, all within a dollar of each other:

That last one breaks the pattern a bit, but I’m including it because it’s the most interesting dark horse in the group.

The question isn’t “which is best.” It’s “which is best for what YOU do.”

ChatGPT Plus ($20/mo): The Generalist King

ChatGPT Plus gives you GPT-5.5 Thinking, 256K reasoning context, agent mode, deep research, Sora video generation (50 clips at 720p), Codex for coding, and 20GB of file storage. I covered the new simplified model picker here if you want to know which model to use when.

GPT-5.5 is legitimately the best general-purpose model out right now. It handles writing, coding, research, image generation, video, and agents without breaking a sweat.

And word on the street is that GPT-5.6 is imminent. Polymarket has it at 89% probability before the end of June. Internal codenames (iris-alpha, ember-alpha, beacon-alpha) have already been spotted in OpenAI’s Codex routing logs.

If you’re subscribing today, you’re buying into an ecosystem that’s about to get another big upgrade for free.

Best for: People who need one AI for everything. Writing, coding, research, image gen, agents, video. If you don’t want to think about which tool to use and just want one that handles most tasks really well, this is it. And with GPT-5.6 around the corner, the value proposition is only getting better.

The catch: If you’re a researcher, Perplexity’s citation game is stronger. If you live in Google Docs, Gemini’s integration is tighter. But for raw capability across the board, GPT-5.5 is hard to beat right now.

Also worth knowing: OpenAI just rolled out ads in the UK for Free and Go tiers. Plus is still ad-free, but the writing’s on the wall for the lower tiers.

Recent changes (May-June 2026): – GPT-5.5 became the default model (May 5) – Simplified model picker with effort-level names instead of model names (June 10) – Memory capacity doubled for Plus/Pro (June 4) – Lockdown Mode for security-conscious users (June 4) – GPT-5.6 rumored for late June (89% on Polymarket, codenames in Codex logs)

Claude Pro ($20/mo): The Specialist Pick

Claude Pro gets you Sonnet 4.6 and Opus 4.8, roughly 45 messages per 5-hour window, Claude Code, 200K context, Projects, and web search. The annual plan drops it to $17/mo ($200 upfront).

Claude’s strong suit is coding. Claude Code is a solid agentic coding tool, and the reasoning quality on complex technical problems is competitive with GPT-5.5. If you’re a developer who’s already in the Claude ecosystem, it’s a fine subscription.

Best for: Developers who prefer Claude’s coding style, writers working with long documents, anyone who needs deep reasoning on technical problems.

The catch: Those message limits are real. About 45 messages per 5 hours sounds like a lot until you’re in a deep session and hit the wall. Heavy users will feel it. And unlike ChatGPT, Claude doesn’t do images, video, or have the same breadth of integrations.

The big change coming June 15: Anthropic is splitting programmatic usage (Agent SDK, CLI, GitHub Actions) out of the flat-rate subscription into a separate metered pool. Your $20 Pro plan now includes a $20 programmatic credit, then you pay API rates beyond that.

Interactive chat is unaffected. But if you’ve been running Claude agents through the SDK on your Pro plan, your costs are about to go up.

Recent changes: – Model retirements on June 15: old Sonnet 4 and Opus 4 IDs being sunset – Claude Fable 5 and Mythos 5 launched June 9 (Mythos-class, Anthropic’s most powerful ever) – Data retention controversy: Mythos-class models require 30-day data retention

Google AI Pro ($19.99/mo): The Bundle Play

Google AI Pro gives you Gemini 3.1 Pro with 1M token context, 5TB of Google storage, YouTube Premium Lite (worth $8.99/mo on its own), the Jules coding agent, and deep integration with Gmail, Docs, and Home devices.

The bundle math is what makes this interesting. If you’re already paying for Google storage and YouTube Premium, the AI features are almost free.

Best for: Google ecosystem people. If you live in Gmail, Google Docs, and YouTube, this subscription pays for itself just in storage and YouTube Premium Lite. The AI is a bonus.

The catch: Google switched from daily prompt limits to a compute-based model that refreshes every 5 hours with a weekly cap. If you exceed it, you get bumped to smaller models. And Gemini still has consistency issues that ChatGPT and Claude don’t. I wrote a guide on getting consistent results if you’re trying to make it work.

Recent changes: – Google I/O 2026 restructure (May 19): new tier system replacing the old Google One AI Premium – New $4.99/mo AI Plus tier (dropped from $7.99) with 400GB storage (June 8) – New $99.99/mo Ultra 5x tier for power users – Top-tier Ultra cut from $250 to $199.99/mo – Pay-as-you-go top-up credits for Pro and Ultra subscribers

If you want the budget option, the new $4.99 AI Plus tier is genuinely compelling. You get Gemini 3.5 Flash, 400GB storage, and NotebookLM access for less than a coffee.

Perplexity Pro ($20/mo): The Research Machine

Perplexity Pro is different from everything else on this list. It’s not a general assistant. It’s a research engine with multi-model access.

You get GPT-5, Claude, Gemini, Kimi, and Nemotron all in one interface. Unlimited searches, 20 deep Research queries per day, the Comet browser, and premium data sources like PitchBook, Statista, and Crunchbase.

Best for: Researchers, analysts, journalists, anyone who needs cited answers from multiple AI models. If your work involves “find me information about X and back it up with sources,” nothing else comes close.

The catch: It’s a search tool, not a general assistant. You can’t build apps with it. You can’t generate images. You can’t have a long back-and-forth conversation like you would with ChatGPT or Claude. It answers questions with citations. That’s its superpower and its limitation.

Recent changes: – Computer in Microsoft 365 launched (May 29): Word, Excel, PowerPoint, Outlook, Teams integration – Personal Computer on Mac available to all users (May 11) – Perplexity announced IPO plans for 2028

I covered getting Perplexity Pro free as a student a while back. If you’re in education, that’s still the best deal in AI.

Copilot Pro ($20/mo): The Office Worker’s AI

Copilot Pro puts AI inside Word, Excel, PowerPoint, and Outlook. You get 100 image generations per day and the new Copilot Cowork feature, which can autonomously work across multiple Microsoft apps.

Best for: Anyone who lives in Microsoft 365. If your job is writing Word docs, building Excel spreadsheets, and answering Outlook emails, this is the one that’ll actually change your daily workflow.

The catch: Copilot Pro technically costs $20/mo, but to get the full Office AI integration, you need Microsoft 365 Personal or Family on top of that ($7-10/mo extra). So you’re really looking at $27-30/mo. Microsoft’s running a $32/mo promotional bundle for July-September 2026, which is honestly the way to go.

Recent changes: – Wave 3 “Copilot Cowork” launched: autonomous multi-app workflows – Agent 365 went GA (May 2026): $15/user/mo add-on for custom agent management – Multi-model support: no longer OpenAI-only, Claude is now integrated

The Dark Horses

SuperGrok ($30/mo)

SuperGrok is the xAI wildcard. You get unlimited queries, deep reasoning with Grok 4, unlimited image and video generation (Aurora), voice mode with cloning, and 128K memory. The $300/mo Heavy tier gets you Grok 4 Heavy, which scored 100% on AIME 2025 and 50.7% on Humanity’s Last Exam.

Best for: X power users, people who need top-tier reasoning, and anyone who wants unlimited AI image and video generation without per-image charges. Grok’s image gen is genuinely impressive and there’s no cap.

The catch: It’s $30, not $20. And the best features (Grok 4 Heavy) are locked behind the $300/mo tier. The base Grok 4 at $30 is solid, but you’re paying a premium for the X integration.

There’s also a $10/mo SuperGrok Lite tier with Grok 3.5 and basic image gen, which is honestly a decent budget option if you want reasoning + image gen without the full price tag.

Poe ($19.99/mo)

Poe is the meta-play. Instead of picking one AI, you get access to GPT, Claude, Gemini, and others through a single interface with compute points. Switch between models on the fly.

Best for: People who want to try everything without committing to multiple subscriptions. If you’re the type who wants Claude for coding, GPT for writing, and Gemini for research, Poe lets you do that for $20 instead of $60.

The catch: You’re getting a middleman version of each model. The experience isn’t as polished as using ChatGPT or Claude directly, and the compute points can run out faster than you’d expect.

The Free Tier Argument

Here’s the thing nobody in the AI industry wants you to think about: you might not need to pay at all.

• Gemini 3.5 Flash is free and genuinely powerful
• ChatGPT Free now includes GPT-5.5 Instant
• NotebookLM is free and absurdly useful for research
• Perplexity Free gives you 5 Pro searches per day
• Grok Free gives you ~50 queries per day

If you’re a casual user, the free tiers have never been better. The paid tiers make sense when you hit specific walls: message limits, context window needs, model quality on hard problems, or specific integrations.

What I’d Actually Recommend

Here’s my honest take, no BS:

If you need one AI for everything: ChatGPT Plus ($20). It’s the best generalist, and GPT-5.5 is the strongest all-around model right now. With GPT-5.6 reportedly dropping this month, you’re about to get even more value.

If you code: ChatGPT Plus ($20). Codex is solid, GPT-5.5 handles coding well, and you get everything else on top. Claude Pro is a reasonable alternative if you specifically prefer its coding style, but you’re giving up a lot of breadth.

If you live in Google: Google AI Pro ($20). The bundle math works out, and the 1M context window is real.

If you research: Perplexity Pro ($20). Multi-model access with citations is a killer combo.

If you’re in Microsoft 365: Copilot Pro ($20 + M365). The Office integration is transformative for spreadsheet people.

If you want two subscriptions: ChatGPT Plus + Perplexity Pro ($40/mo total). General-purpose powerhouse plus a research engine. That combo covers 95% of use cases.

If you’re broke: The free tiers are better than ever. Start there. Upgrade when you hit a wall.

What’s Changing

The AI subscription landscape shifted in the last 30 days, and it’s not done shifting.

GPT-5.6 is coming. Polymarket has it at 89% probability before the end of June. Internal codenames have been spotted in OpenAI’s Codex logs. If you’re on the fence about subscribing to ChatGPT Plus, the timing is about as good as it gets. You’re buying into an ecosystem that’s about to get a major upgrade.

Claude’s June 15 repricing is the other big one. If you use Claude’s Agent SDK or CLI on your Pro plan, your costs are about to go up. Programmatic usage is being metered separately with monthly credits. Interactive chat is unaffected, but this is Anthropic quietly acknowledging that flat-rate agent usage isn’t sustainable.

Google’s $4.99 AI Plus tier just dropped. It’s the cheapest serious AI subscription available, and it comes with 400GB of storage. If budget matters, this is the move.

The agent usage split is a trend, not a one-off. As AI agents become more capable and more compute-hungry, expect every provider to separate “chat with AI” from “AI does work on your behalf” pricing. The days of unlimited agent usage on a flat $20/mo plan are ending.

Pick the subscription that matches your workflow. If you’re paying for one you don’t use, cancel it today and try the one you actually need. And if you’re only going to pay for one, ChatGPT Plus is the safest bet right now. It’s about to get even better.