Which AI coding plan is worth it at $20, $100, and $200 a month (October 2026)

We re-read every price below on the vendor's own page on October 14, 2026. US prices, before tax. None of these vendors prints a token or message cap for its coding allowance. So this guide uses two kinds of evidence: what the pages promise, and what developers on X report after hitting the limits. The X posts are personal experiences, not measurements. We grouped them where several people describe the same thing, and we label single reports as single reports. We did not run a usage test.

How the limits work, in one paragraph each

• Claude limits reset on a rolling five-hour window, and paid plans add weekly limits on top. Chat, Claude Code, and the desktop apps share one pool. Max is billed monthly only.
• ChatGPT Pro has no five-hour window in Codex and Work. OpenAI's Codex lead committed to "not reintroducing the 5h limit" on September 29. The weekly allowance is what runs out. The Pro tiers article prints no cap.
• The X Subscription has one pool shared across Grok and Cursor that refreshes weekly. Third-party models in Cursor are billed separately at API list prices.
• GitHub Copilot sells AI credits worth $0.01 each. Completions and next-edit suggestions don't use credits on paid plans.
• Google AI app limits refresh every 5 hours until a weekly limit, and can be extended with AI credits.

Around $20 a month

Plan	Monthly price	What the page says about usage
Claude Pro	$20, or $17/month billed $200 a year	At least 5x Free per five-hour session, plus weekly limits. Claude Code included.
ChatGPT Plus	Not published on the page we read	"Expanded Codex usage." GPT-6.1 Sol, GPT-6 Astra, and GPT-6 Sol listed.
Google AI Pro	$19.99 in our pricing table. The page we read on Oct 14 did not render it.	4x Free. Gemini 4 Argon in the app. "Entry rate limits" in Antigravity.
GitHub Copilot Pro	$10	1,500 AI credits a month (1,000 base plus 500 flex)
GitHub Copilot Pro+	$39	7,000 AI credits (3,900 base plus 3,100 flex). Premium models.
X Subscription Plus / Premium	$8 / $30	Shared weekly Grok-and-Cursor pool. Premium is 4x Plus and adds Grok Bot.
Legacy Cursor Pro (existing subscribers)	$20	A Cursor Models pool and an Other Models pool, reset monthly

Who it suits. People who code in sessions rather than all day, and anyone testing which tool fits before committing $100 or more.

What developers report. The most consistent Claude Pro report since Opus 5.5 is that the five-hour window binds before the weekly cap does. One developer working 8–10 hour days wrote: "The bottleneck is the 5-hour window, not the weekly cap. Two sessions a day don't reach 1/7 of the weekly quota." Another measured that "Sonnet 5.5 on xhigh burned 32.3M tokens in 42 minutes on Claude Pro. That used 6% of the weekly limit and 49% of the 5 hour limit." Effort level matters a lot. The same plan feels roomy on medium and tight on xhigh.

On Copilot, several developers describe running out of credits before the month ends, usually after switching to a premium model or running subagents. Some examples: "I ran out of copilot credits last night. Back to coding by hand like a peasant". "I forgot the love for different models in subagents. Instantly ran out of copilot credits." One developer said a single planning prompt with GPT-6.1 Sol "burned through 125 credits". At GitHub's $0.01 rate, that is $1.25 of a $10 plan's 1,500 credits.

Hidden costs.

• Copilot overage is billed per credit, and the model you pick sets how fast credits go.
• ChatGPT Plus has no printed price on OpenAI's page, so check checkout before you budget.
• On the X Subscription, any Claude or GPT use in Cursor is extra from the first request (x.ai/pricing).

Bottom line. At $20, Claude Pro is the only plan in this group with a published five-hour-plus-weekly structure and Claude Code included. Users report the five-hour window as the real limit. Copilot Pro is the cheapest way to get unlimited completions, but agent work spends credits fast. The X Subscription has no $20 tier: Plus is $8 and Premium is $30. Neither covers third-party models in Cursor.

Around $100 a month

Plan	Monthly price	What the page says about usage
X Subscription Super	$100	15x Plus, priority access at peak times
Claude Max 5x	"From $100"	5x Pro per five-hour session. $100/month in API credits.
ChatGPT Pro 100	$100	Ultrafast not included. No cap printed. Monthly only.
GitHub Copilot Max	$100	20,000 AI credits (10,000 base plus 10,000 flex)
Google AI Ultra, 5x	$99.99	5x AI Pro. "Higher rate limits" in Antigravity.

Who it suits. People who use an agent daily and hit the $20 limits most days.

What developers report. The tier most often named as the value pick on X is Claude Max 5x, mainly because of how the 20x tier's weekly cap behaves (see below). Several developers who dropped ChatGPT Pro 200 after the allowance change described pairing two $100 plans instead. One wrote "Well, goodbye ChatGPT Pro 200 / Hello ChatGPT Pro 100 and Claude Max 5x". Another wrote "Switching to ChatGPT Pro 100 and Claude Max 5x seems to give best of both worlds."

Heavy Fable use is the most common way to run out. A Max user reported hitting the five-hour limit "after about 34 minutes and 34% of my Fable weekly limit" on a single research prompt. Claude's own table caps Fable at 50% of weekly limits on Max.

Hidden costs. Claude Max is monthly only. ChatGPT Pro 100 does not include Ultrafast, and buying credits does not unlock it (OpenAI). X Super, like every X tier, bills Claude and GPT in Cursor separately.

Bottom line. For one-vendor heavy use, the $100 tier is where most people in our sample report the limits stop getting in the way. Claude Max 5x has the most positive reports. ChatGPT Pro 100 is the cheaper OpenAI route now that Pro 200's allowance is lower. X Super only makes sense if your work is mostly Grok.

Around $200 a month

Plan	Monthly price	What the page says about usage
X Subscription Ultra	$200	50x Plus
Claude Max 20x	Not printed separately ("From $100")	20x Pro per five-hour session. $200/month in API credits.
ChatGPT Pro 200	$200	More than Pro 100. New subscriptions get a lower allowance than the old Pro 200.
Google AI Ultra, 20x	$199.99	20x AI Pro. "Highest rate limits" in Antigravity.
Legacy Cursor Ultra (existing subscribers)	$200	Both pools, numeric cap not printed

Who it suits. People running several agents in parallel, or long autonomous sessions.

What developers report.

Claude Max 20x's "20x" applies to the five-hour window, not the week. Several developers say the weekly cap on Max 20x is far less than four times Max 5x. "The $200/mo Claude Max so-called "20x" plan gives you a higher 5-hour limit, but your weekly limit is exactly the same as the $100/mo plan," one wrote. Others put it at about 1.5x. "We basically pay 2× the price but only get 1.5× the tokens." "You pay $200 instead of $100 to get like 1.5x more usage out of it." These are user measurements. Anthropic's page describes 20x only "per 5-hour session" and does not print a weekly multiple.

Even so, heavy users report Max 20x lasting. "I've been running 2 Opus 5.5 autonomous sessions for almost a week… Claude Max x20 still feels like much better value for this kind of work". Another user ran "Opus 5.5 Extra in Claude Code on a single task for 7.5 HOURS straight. And somehow, it's consumed just 5% of my 5-hour usage limit". Not everyone agrees: "Just burnt 16% of my Claude Max weekly limit in just 37% of my 5 hour limit".

ChatGPT Pro 200 drains fast on Astra. Several Pro 200 subscribers describe using up the weekly allowance in one to two days on GPT-6 Astra. One wrote: "I've been exhausting the weekly limit in 2 days (Astra)". Another hit the weekly limit, slipped into paid credits, and logged "35,000 credits in past ~day (it would cost $1,400 to buy back 35,000 credits)". The piece on the 28-day run and Pro 200 covers this tier in depth.

Cursor's legacy plans and third-party models. Before The X Subscription, one developer reported "That weekend there I spent $150 on On Demand usage alone" running Opus 5.5 in Cursor, and moved to Claude Code. Since the X Subscription launched, developers on legacy Cursor Ultra say the Other Models pool is the reason to stay. "If I merge I lose $400 worth of Claude, GPT (for now), Gemini and others… No way I'm giving that up," one wrote. Another said they are "staying on Pro+ until Grok catches up to Opus 5.5 for my work." The $400 figure is that user's own reading. Cursor's docs don't print the pool size.

Hidden costs.

• On the X Subscription, Claude and GPT in Cursor are billed at API rates from the first call, with no included allowance (x.ai/pricing).
• ChatGPT Pro 200 subscribers who joined before the change move to the lower allowance after October 29 (OpenAI).
• On Claude, Fable is limited to 50% of weekly limits on Max.

Bottom line. At $200, what matters is how each vendor counts the week, not the sticker price. In our sample, Claude Max 20x gets the strongest "it lasts" reports, but several users say its weekly cap is about 1.5x Max 5x, not 4x. That makes Max 5x the better value unless you keep hitting the five-hour window. ChatGPT Pro 200 now buys half the old allowance. X Ultra is the largest pool here for Grok models and includes nothing for Claude or GPT. Legacy Cursor Ultra holders who rely on Claude in Cursor have a reason not to switch.

What we cannot tell you

No vendor publishes token counts, so the posts above are the closest thing to a measurement. They come from people who chose to post, and heavy users are over-represented. Treat them as a guide to which questions to ask, not as a promise of what you will get.

The full table, with sources for every row, is on AI Model & Plan Prices. Related: Before you switch a Cursor seat onto The X Subscription and ChatGPT Pro's three prices, and Claude Max from $100.