Choosing an AI subscription means weighing a flat monthly fee against pay-per-use API access. This article compares the consumer tiers of ChatGPT, Claude, Gemini, Microsoft 365 Copilot, and Perplexity, then explains the pricing models behind them. You are reading this to find the plan that matches how you actually use AI.

Prices and tiers change often. Every figure here is stated as of June 2026 and links the official pricing page. Check the linked page before you buy. You are a European audience, so euro figures appear where the official EU page lists them in euros, and US dollar figures appear where the official page showed US dollars.

A transparent chest filling with golden energy on a dark background. No em-dashes.
Deciding on an AI subscription is about judging how much value pours back out for the fee you pour in.

A quick decision aid

Start with how often you use AI and what you are building. The flow below points you to a tier class.

Path A Occasional individual A few questions per week. Stay on a free tier or a low entry tier.
→
Path B Daily power user Hours every day, long documents, heavy chat. Move to a Plus or Pro class tier.
→
Path C Product builder Calling the model from code at scale. Use usage-based API access, not a seat subscription.

Consumer subscription tiers at a glance

The table below lists the main consumer tiers for each vendor. Currency matches the official page exactly. Prices are as of June 2026.

VendorTierPrice (as of June 2026)What you get
ChatGPTOpenAIFree0 €/monthBasic access
ChatGPTOpenAIGo7,99 €/monthEntry paid access
ChatGPTOpenAIPlus23 €/monthHigher limits, more models
ChatGPTOpenAIPro103 €/monthHighest consumer limits
ClaudeAnthropicFree$0Basic access
ClaudeAnthropicPro$17/mo annual, $20/mo monthlyHigher usage limits
ClaudeAnthropicMaxfrom $100/month5x or 20x Pro usage
GeminiGoogleFree0 €/monthBasic access
GeminiGoogleAI Plus4,99 €/monthEntry paid access
GeminiGoogleAI Pro21,99 €/monthHigher limits, more credits
GeminiGoogleAI Ultra99,99 € or 219,99 €/monthTwo Ultra tiers, most credits
CopilotMicrosoftM365 Personal$9.99/mo or $99.99/yearOffice apps plus Copilot
CopilotMicrosoftM365 Family$12.99/mo or $129.99/yearUp to six people
CopilotMicrosoftM365 Premium$19.99/mo or $199.99/yearAI agents, research reports
PerplexityPerplexityFree$0Basic search
PerplexityPerplexityPro$20/monthHigher limits, more models

Consumer subscription tiers explained

OpenAI ChatGPT

ChatGPT runs from a Free tier at 0 €/month up to Pro at 103 €/month, with Go at 7,99 €/month and Plus at 23 €/month in between (as of June 2026, chatgpt.com/pricing ). The EU pricing page showed these figures in euros. The Pro tier shown here is the lower of OpenAI’s two Pro tiers. OpenAI’s Help Center confirms Pro has two tiers, at $100 and $200 US dollars, described at help.openai.com . For teams, Business costs 21 €/user/month on annual billing or 26 €/user/month on monthly billing. Enterprise pricing is custom.

Anthropic Claude

Claude starts at Free ($0), then Pro at $17/month on annual billing or $20/month on monthly billing (as of June 2026, claude.com/pricing ). The Claude pricing page showed US dollars. Max starts from $100/month and gives 5x or 20x Pro usage. For teams, the standard seat costs $20/seat/month annual or $25 monthly, for 5 to 150 seats. The Team premium seat costs $100/seat/month on annual billing. Enterprise is $20/seat plus usage at API rates, with custom terms.

Google Gemini

Google’s plans run from Free (0 €/month) through Google AI Plus at 4,99 €/month and Google AI Pro at 21,99 €/month, up to Google AI Ultra (as of June 2026, gemini.google/subscriptions ). The EU page showed euros. Ultra has a lower tier at 99,99 €/month and a higher tier at 219,99 €/month. Google AI plans meter “Flow Credits”, a pool of credits spent on video and agent generation: 200, 1.000, or 10.000 to 25.000 credits depending on the tier.

Microsoft 365 Copilot

Microsoft folds consumer Copilot into Microsoft 365. Personal costs $9.99/month or $99.99/year, Family costs $12.99/month or $129.99/year, and Premium costs $19.99/month or $199.99/year (as of June 2026, microsoft.com ). This page showed US dollars. Premium adds AI agents, research reports, and extensive Copilot use. The standalone “Copilot Pro” plan was retired and folded into Microsoft 365 Premium, so it is not a current tier.

Perplexity

Perplexity offers a Free tier ($0) and Pro at $20/month (as of June 2026, perplexity.ai help center ). Perplexity also offers a Max tier, announced on the official blog , but its price is not confirmed on the official pricing page as of June 2026, so this article does not state a Max price.

Subscription versus API access

A consumer or team subscription is a per-seat flat fee. You pay a fixed amount per user each month, and the vendor caps your usage instead of metering it (sources: claude.com/pricing , chatgpt.com/pricing ). You log in to a chat app and use it through a browser.

API access is different. You call the model from your own code, and you pay per token. A token is a chunk of text, roughly a few characters. You are billed per million tokens, and input and output are priced separately. Output costs more than input. Subscriptions suit people who chat with the model. API access suits builders who run the model inside a product.

Subscription
Flat monthly fee Per seat Usage is capped, not metered. Use through a chat app.
API
Pay per token Input and output priced separately Output costs more than input. Call from your own code.

The pricing models behind the tiers

Five pricing models cover almost every plan on this page.

Per-seat flat subscription. A fixed monthly fee per user. The vendor caps how much you can use rather than billing each request. This covers all consumer and team tiers above (sources: claude.com/pricing , chatgpt.com/pricing ).

Per-token usage-based. API billing per million tokens, input and output separate, output more expensive. Official example figures: the Anthropic Opus class costs $5 input and $25 output per million tokens, the Sonnet class costs $3 and $15, and the Haiku class costs $1 and $5 (Anthropic API pricing ). OpenAI publishes its own per-token rates (OpenAI API pricing ).

Prompt caching discount. When you reuse the same context across calls, the cached portion is billed far cheaper. OpenAI prices cached input at about 10% of standard input. Anthropic prices cache reads at about 0.1x the base input rate, a roughly 90% discount, while cache writes cost more upfront (sources as above).

Batch API discount. Both OpenAI and Anthropic process bulk jobs asynchronously at 50% off input and output. Use this when you do not need an instant answer (sources as above).

Hybrid per-seat plus usage. Enterprise plans add metered consumption on top of a seat license. Anthropic Enterprise is $20/seat plus usage at API rates (claude.com/pricing ).

Credit-based hybrid. Some consumer power tiers bundle a pool of generation credits into the flat fee. Google AI plans include Flow Credits for video and agent generation (gemini.google/subscriptions ).

How to choose

Occasional user. Use a free tier. If limits annoy you, move to a low entry tier such as ChatGPT Go (7,99 €/month) or Google AI Plus (4,99 €/month), as of June 2026.

Power user. A Plus or Pro class subscription gives predictable cost and high limits. ChatGPT Plus (23 €/month), Claude Pro ($17/month annual), and Google AI Pro (21,99 €/month) all sit in this band, as of June 2026.

Team. Buy per-seat team plans for shared admin and billing. Claude Team standard is $20/seat/month annual, and ChatGPT Business is 21 €/user/month annual, as of June 2026.

Product builder. Use the API and pay per token. Apply prompt caching for repeated context and the Batch API for non-urgent jobs to cut cost. A flat subscription does not scale across many automated calls.

Worked example: subscription versus API tokens

This is an illustration, not a quote. It shows when a flat subscription beats per-token API billing and when it loses. Figures use the verified Anthropic Sonnet class example: $3 per million input tokens and $15 per million output tokens.

text
Illustration only. Figures as of June 2026.

Sample workload: 1,000 model calls in a month.
Each call: 2,000 input tokens, 500 output tokens.

Totals across 1,000 calls:
  input  = 1,000 x 2,000   = 2,000,000 tokens = 2.0 million
  output = 1,000 x   500   =   500,000 tokens = 0.5 million

Option 1: API (Sonnet class example, $3 input / $15 output per million)
  input cost  = 2.0 x $3  = $6.00
  output cost = 0.5 x $15 = $7.50
  API total   = $13.50 per month

Option 2: Flat subscription
  Claude Pro  = $20 per month (monthly billing)

Result for THIS workload:
  API ($13.50) is cheaper than the subscription ($20).

Now scale up 10x: 10,000 calls in the month.
  input  = 20 million x $3  = $60
  output =  5 million x $15 = $75
  API total = $135 per month, far above any single seat fee.

Takeaway:
  Light, code-driven workloads can be cheaper on the API.
  Heavy automated volume pushes API cost above a flat seat.
  A human chatting all day is usually cheapest on a flat subscription.

The lesson is about volume and shape. Low automated volume can favour the API. High volume favours either heavy caching and batching or a flat plan. A person chatting through a browser nearly always pays less on a subscription, because a subscription caps usage instead of metering every token.

Further reading