跳到正文
DEV Community· ai-saasradar·· 26 分鐘前精選AI 評分76

2026 年 ChatGPT Plus 與 Claude Pro 月費近 $20,功能權限各異

ChatGPT and Claude both sell a seat near $20 in 2026, and they gate completely different things

AI 導讀

按作者於 2026 年 11 月 22 日查看的定價頁換算,ChatGPT Plus 每月約 $19.63,Claude Pro 每月 $20,但兩者解鎖的功能和方案限制不同。

推薦理由

文章將兩個約 $20 方案在模型權限、編碼智能體、上下文資訊及年付價格上的差異並列,可按工作需求和付款週期作具體比較。

正文 · 原文

ChatGPT and Claude both sell a seat near $20 in 2026 — and they gate completely different things

Two pricing pages, opened within a minute of each other on 22 November 2026, both pointing at a seat around twenty dollars. OpenAI served me won; Anthropic served me dollars. Once you remove the Korean VAT and convert, ChatGPT Plus is $19.63 a month at 1,343.33 KRW/USD and Claude Pro is $20. That is where the agreement ends.

What follows is the two ladders as printed, then the four places where the $20 seat is not the same purchase: the model gate, the context budget, the annual toggle, and the free tier.

The ladders as printed

Plan Price What the page says changes
ChatGPT Free ₩0 Unlimited text chats with GPT-6; limited uploads, images, voice, deep research, memory, Codex
ChatGPT Go ₩13,000/mo (~$8.80 ex-VAT) More messages, uploads, images, voice; longer memory; plugins; scheduled tasks; may include ads
ChatGPT Plus ₩29,000/mo (~$19.63 ex-VAT) GPT-6 reasoning models, expanded limits, projects, scheduled tasks, custom GPTs, expanded Codex
ChatGPT Pro from ₩159,000/mo (~$107.60 ex-VAT) A choice of 3 usage tiers (Pro 100/200/500 at ₩159,000/299,000/790,000), GPT-6 Astra reasoning, the always-on agent "Dot", unlimited images, API credits
Claude Free $0 Limited usage every 5 hours and every week
Claude Pro $20/mo, or $17/mo with $200 billed up front Claude Code, task hand-off and scheduling, Projects, more models, Claude in Chrome and Microsoft 365
Claude Max $100/mo 5× Pro usage, $100 a month in API credits, Fable included up to half the weekly limit
Claude Ultra $200/mo 10× Pro usage, no 5-hour limit for verified accounts, $200 a month in API credits
Claude Team $25/seat/mo, or $20 annually 2 to 150 seats; a premium seat at $125 monthly / $100 annually carries 5× the usage

Seat prices by plan

Read the two middle rows together and the shapes are different. Claude jumps $20 → $100 → $200 in steps of 5× and 10× usage. ChatGPT jumps $20 → $107.60 with a different promise: the frontier reasoning model, an always-on agent, image generation without a cap, and a monthly API credit allowance. One ladder sells more of the same thing; the other sells a class of thing you cannot buy below it.

The $20 seat: what it actually unlocks

On Claude's side, Pro is the first plan that includes Claude Code — the coding agent — plus the ability to hand off and schedule tasks. The feature table marks both "No" on Free and "Yes" on Pro. It also unlocks Projects and the Microsoft 365 and Chrome integrations.

On ChatGPT's side, Plus is the first plan that includes GPT-6 Astra and GPT-6.1 Sol, the frontier models. The comparison table is explicit: Free and Go are "No" on both; Plus is "Limited"; Pro is "Expanded". Plus also brings projects, scheduled tasks and custom GPTs — though Go already gets scheduled tasks, which is a quiet difference from last year's ladder.

So the same twenty dollars buys a coding agent on one side and access to the frontier chat models on the other. If your work is "an agent that writes and runs code", Claude Pro is the only one of the two $20 seats that includes it. If your work is "the best general model for writing and analysis", ChatGPT Plus is the only one of the two $20 seats that includes the top model at all.

The always-on agent is a third tier on both sides and costs differently: ChatGPT's "Dot" starts at Pro, ₩159,000 (~$107.60), and Claude's scheduling arrives at Pro, $20.

Only one vendor prints a context budget

Context window is the number that decides whether a job fits. OpenAI prints it per plan:

  • GPT Instant total context: 27K on Free, 54K on Go and Plus, 128K on Pro
  • GPT reasoning total context: 256K on Go and Plus, 400K on Pro
  • Input ceiling per message, in the page's own units: about 12 pages on Free, 40 on Go and Plus, 250 on Pro

Context window by plan

Anthropic's pricing page, read the same day, prints no context figure at any tier. It prints usage as a multiple ("more usage than Free", "5× Pro usage", "10× Pro usage") and leaves the token budget to its documentation. That is not a criticism of the product, but it is a real difference in what you can plan from the pricing page: with ChatGPT you can calculate whether a 200-page document fits on the plan you are looking at. With Claude you cannot, from this page.

The annual toggle exists on exactly one of them

Claude Pro is $20 billed monthly or $17 a month with $200 billed up front — a 16.7% discount for committing a year, stated on the card.

ChatGPT's own FAQ says: "We offer monthly plans for Go, Plus and Business and annual plans for Business and Enterprise." There is no annual individual plan to toggle. So the annual lever is worth $40 a year on the Claude side and does not exist on the ChatGPT side at this tier.

Twelve months of Pro, monthly against annual

At ten seats the lever gets bigger. Claude Team standard seats are $25 monthly or $20 annually, so ten seats run $250 a month billed monthly or $200 a month on the annual commitment — $600 a year apart. For ChatGPT's Business tier I did not read a printed per-seat price on the page that served me, so I am not quoting one; that figure has to come from the Business tab or a sales conversation.

The free tiers are not the same fork in the road

Claude's Free plan is a volume limit: "limited usage every 5 hours and every week", with chat, search, file creation, code running, memory and Artifacts included. You get the tools and a small tank.

ChatGPT's Free plan is a model gate: unlimited text chats with GPT-6 — but "No" on GPT-6.1 Sol, GPT-6 Astra and GPT-6 Sol, with GPT-6 Luna available. You get a large tank of the smaller model.

That difference matters more than the price. Someone testing whether an assistant can handle their week should look at how much they can use it; someone testing whether the frontier model is worth paying for gets nothing from ChatGPT's free tier, because the frontier model is not on it.

What neither page tells you

  1. What a "usage tier" is. ChatGPT Pro's three tiers are named Pro 100, Pro 200 and Pro 500 — and the page's own footnote explains the numbers are API credit allowances (₩159,000 / ₩299,000 / ₩790,000, with ₩790,000 of API credits on the top one), not a published message count. Claude's "5× Pro usage" and "10× Pro usage" are equally undefined in absolute terms.
  2. What "unlimited" excludes. Both pages put abuse guardrails on unlimited text chats, in a footnote rather than the card. Neither publishes a throttle number.
  3. Tax. The won figures on OpenAI's Korean page include 10% VAT. The dollar figures on Claude's page do not include tax. A like-for-like comparison has to add Korean VAT back and subtract it from the other side, which is what the conversion above does.
  4. Whether the model you want is on your plan at all. Both vendors now gate specific models by tier rather than selling "the model". On ChatGPT that is a per-model row across four plans; on Claude it is "more Claude models" on the Pro card plus Fable metered by usage credits.

If you are deciding this week

  • You want a coding agent for $20: Claude Pro. It is the only $20 seat of the two that includes one.
  • You want the frontier chat model for $20: ChatGPT Plus. Free and Go both say "No" to GPT-6 Astra and GPT-6.1 Sol.
  • You want a planning number for context: ChatGPT, because 27K/54K/128K is printed. If context is your constraint on the Claude side, the pricing page will not answer it.
  • You are buying a year: Claude Pro at $200 up front is $40 cheaper than twelve months at $20. There is no equivalent on the ChatGPT individual side.
  • You are buying ten seats: Claude Team's annual price ($20/seat) is $600 a year below its monthly price on ten seats. Get ChatGPT's Business quote in writing before comparing, because it is not on the public individual page.

The pair is laid out with both entry prices and their verification dates on ChatGPT vs Claude. The single-tool pages for ChatGPT and Claude carry their own records, and if a third option is in play, ChatGPT vs Gemini and Claude vs Gemini use the same method. The full lists are at ChatGPT alternatives and Claude alternatives.

Prices were read from chatgpt.com's pricing page (served in Korean won, VAT included) and claude.com/pricing (US dollars, tax excluded) on 22 November 2026. Won prices were converted by removing 10% VAT and dividing by 1,343.33 KRW/USD, the rate published by open.er-api.com at 00:02 UTC that day; the dollar figures that result are arithmetic, not the price a US buyer sees. Usage multiples, "unlimited" and per-model gating are quoted as the pages word them. Anything a page did not state — a per-seat Business price, an absolute token budget on Claude's side, a throttle number — is marked as not stated rather than estimated.

來源:DEV Community · dev.to