How much does AI actually cost in 2026: every major plan, checked and dated

Every consumer plan, side by side#

The market has converged on three numbers#

This is the single most useful thing to understand about AI pricing this year.

Anthropic set the $100 and $200 rungs first with Max 5x and Max 20x. OpenAI matched with the Pro tiers. Google matched at I/O 2026 and adopted the same 5x and 20x multiplier language.

What that means for you: price is no longer how you choose. When four companies charge identical amounts, the decision moves entirely to capability, ecosystem and what else is bundled.

The two that break the pattern#

Grok goes higher. SuperGrok is $30 for the standard tier — 50 percent above everyone else — and Heavy is $300, the most expensive consumer AI subscription on the market. Nobody else goes past $200.

DeepSeek does not play. There is no subscription at any price. Web chat is free with no paid tier above it, and the API is per token. It is the only major provider with nothing to cancel.

What each standard plan includes for its ~$20#

This is where the identical prices stop being identical.

Google and Microsoft are selling bundles. OpenAI and Anthropic are selling the assistant alone. Perplexity is selling a narrower tool with published caps.

The cheapest ways to pay#

That last row is the most overlooked thing in this market. Microsoft moved Copilot features into the base Microsoft 365 Personal and Family plans and raised those plans by roughly $3 a month. If you already pay for Office as an individual, you have AI access for three dollars and no separate subscription.

API pricing is a completely different scale#

Subscriptions cluster within a few dollars of each other. Token pricing does not.

These are not equivalent models and we are not claiming they are. GPT-6 Astra has a 1,050,000-token context window and frontier reasoning that the cheaper models do not match.

But a 45-fold gap is large enough that "we default to the best model" deserves to be a decision rather than a habit. For classification, extraction and summarisation, the cheap models are frequently good enough. For hard reasoning they frequently are not. The only way to know is to run your own workload against both.

Three figures we could not settle#

Our methodology is to report disagreement rather than resolve it by picking a favourite. Three prices on this page are genuinely unclear:

In all three cases: confirm on the provider's own page before you commit money.

One price has a deadline#

How to actually decide#

Ignore the model benchmarks for a moment and answer three questions in order.

1. Where do your files live?

Neither model advantage is worth migrating an entire business between office suites.

2. What is your budget?

3. What is the work?

The mistake almost everyone makes#

Paying for two.

At $20 a month each, running ChatGPT and Claude, or Gemini and Perplexity, costs $480 a year for heavily overlapping tools. Pick the one that matches your main job and use a competitor's free tier when you want a second opinion.

The free tiers in 2026 are genuinely good. That is the quiet story in all of this: for a large share of small businesses, the correct answer is still $0, and the honest test is to run a fortnight on free and count how often you actually hit a wall.