Two ways to read this: browse everything at a price you are willing to pay, or put two plans side by side. Either way you get our verdict on the plan, what it is good and bad at, and the evidence behind it. Want a recommendation instead? Take the quiz. For agent-against-agent results, compare models.
Is one plan worth it on its own? See the plan pages.
Showing the standard band. Every band is on this page - pick another above, or scroll.
Free $0/mo
Our pick: Cursor Hobby (free). Cursor Hobby is the only free plan with a real agent that edits your code. ChatGPT Free only lets you chat about code. Use Hobby to find out whether you want an agent in your editor at all. Do not judge the models by it. Composer 2.5 is weak, and the requests run out after a few tasks.
Most free tiers are chat with a demo attached. This one is a real agent editing your repo. The catch sits at both ends. Composer 2.5 is a weak model, and the requests run out within a few tasks. Use it to find out whether you like the workflow, not whether you like the model.
Pros
An agent that edits your repo, for nothing
The same agent and IDE as Cursor Pro
No card, nothing to cancel
Cons
A few tasks and the requests are gone
Composer 2.5 is not capable enough for serious work
OpenAI plans to leave Cursor on November 12, 2026
Best for Finding out if you want an agent in your editor at all.
How far it goes: we have not measured this plan
We grade the same Cursor agent on the $20 Pro plan - its scorecard is the closest evidence.
A capable chat app with a taste of Codex on the side. Ask it questions all day. Do not plan on it carrying work to the end, because the Codex allowance is tiny. If you want to try Codex properly, ChatGPT Plus at $20 is where to start.
Pros
Costs nothing
Fine for questions and drafts
Cons
Very little Codex usage
Real agent work needs a paid plan
Best for Occasional questions and drafting.
How far it goes: we have not measured this plan
Same Codex CLI family as the paid ChatGPT plans with tight limits - the ChatGPT Plus scorecards are the closest evidence.
Our pick: ChatGPT Go. Nothing under $20 buys enough agent usage for real work. Our advice is to start at $20 instead. If you do want a cheaper first step, Go is the one that leads somewhere: it runs the same Codex CLI as ChatGPT Plus, so the upgrade changes only the limit. Google AI Plus and SuperGrok Lite have no coding agent worth the name.
Go gets you onto Codex for $8. It runs the same Codex CLI as ChatGPT Plus, so nothing changes when you upgrade except the limit. The limit is what you will notice first. We would rather start at $20, but if $8 is the budget, this is the plan.
Pros
The cheapest way onto Codex
Same Codex CLI as Plus
Cons
Tight enough that you will soon want Plus
Not enough usage for real work
Best for A cheap first step onto Codex.
How far it goes: we have not measured this plan
Same Codex CLI family as ChatGPT Plus with tighter limits - the Plus scorecards are the closest evidence.
$10 buys longer Grok chats and limited image and video generation. It does not include a coding agent. This plan only makes sense if you already prefer Grok and regularly hit the free limits.
Pros
More Grok usage than free
Basic image and video generation
Cons
No coding agent
Limited value over the free plan
Best for People who already use Grok chat and hit the free limits.
How far it goes: we have not measured this plan
We grade Grok CLI on the $100 SuperGrok Plus plan - its scorecard is the closest evidence.
Our pick: Claude Pro. Best starter pick. Opus 5.5 is the strongest model we use day to day and it is available on the $20 Claude plan. It is light on the limit, so $20 goes a long way. Sonnet 5.5 on the same plan handles smaller things and simple tasks. Just pick Claude Pro if you are unsure what you want. ChatGPT Plus is the pick if you want Codex, with Luna for small tasks when quota matters most. Google AI Pro makes sense only when its bundle matters. Cursor Pro is for people who want the agent inside the IDE, but OpenAI plans to leave Cursor on November 12, 2026. We do not recommend SuperGrok or Kimi.
Claude Pro includes Opus 5.5 through Claude Code, the best CLI agent. Opus 5.5 does most of what Fable does on far less of the limit. That leaves little reason to pay for Max just to get Fable. Sonnet 5.5 is now the model for smaller things and simple tasks.
Pros
Opus 5.5 is really good and you can use it a fair bit on the $20 plan
Sonnet 5.5 for smaller things and simple tasks
Claude Code is the best CLI agent
Decent usage limits
Cons
Fable requires the $100 Max plan
Best for People who want the best coding subscription at $20.
How far it goes: we have not measured this plan
The Max 5x scorecards are the closest evidence, minus Fable: that model is Max-only, so the best agent this plan can run is Opus 5.5.
Kimi K3 can produce good-looking interfaces, but it does less with the same prompts than it did a few months ago. Moderato's credits run out before ordinary work is done. We do not recommend it.
Pros
Good visual design
Long context
Cons
Credits drain too quickly
K3 has regressed in our use
Best for Nobody.
How far it goes: we have not measured this plan
No graded Kimi subject at the moment; see the leaderboard for graded alternatives.
$20 gets you Codex with every OpenAI model, including Astra. Luna handles high-volume small tasks at a fraction of the quota. OpenAI's extra resets are getting fewer and fewer, so don't rely on them lasting much longer.
Pros
Every OpenAI model, including Astra
Luna is cheap on quota
Generous allowance while the extra resets last
Cons
Opus 5.5 is stronger at the same price
Astra can drain the whole 5-hour window in one prompt
The current allowance depends on temporary extra resets
Best for People who want Luna for high-volume small tasks.
How far it goes: at least 15 times a week, and more than 100 for the cheapest model
Our whole real-world issues and spec planning pass runs at least 15 times a week, and more than 100 for the cheapest model on this plan, depending on which model you run.
That works out at about under 5c to 28c of the subscription per run, at the list price.
Measured on this plan, at <1% to 6%/1w of its allowance per run. A share belongs to the plan it was measured on, so we never subtract one plan's from another's.
Cursor Pro lets you switch between providers. Its agent runs inside the editor, which suits people who still read and edit the code themselves. Composer 2.5 is not capable enough for serious work. OpenAI plans to leave Cursor on November 12, 2026, which will narrow the model choice.
Pros
Agent built into the IDE
Models from several providers
Cons
Composer 2.5 is weak
OpenAI plans to leave Cursor on November 12, 2026
Best for People who still read their code and want to switch between models.
How far it goes: about 50 times a month
Our whole real-world issues and spec planning pass runs about 50 times a month on this plan.
That works out at about 41c of the subscription per run, at the list price.
Measured on this plan, at 2.0%/1m of its allowance per run. A share belongs to the plan it was measured on, so we never subtract one plan's from another's.
Google AI Pro has no Gemini model we would pick for coding now. Sonnet 5.5 does the fast, cheap work better. Antigravity is rough and often leaves you guessing what it is doing. The plan also includes 5 TB of storage, NotebookLM, YouTube Premium Lite, Home Premium and Gemini in Google apps.
Pros
5 TB of storage
A broad Google services bundle
Cons
No Gemini model we would pick for coding
Antigravity is rough
Best for People who want basic AI usage and will use the Google bundle.
How far it goes: about 4 to 6 times a week
Our whole real-world issues and spec planning pass runs about 4 to 6 times a week on this plan, depending on which model you run.
That works out at about 73c to $1.09 of the subscription per run, at the list price.
Measured on this plan, at 15.9% to 23.7%/1w of its allowance per run. A share belongs to the plan it was measured on, so we never subtract one plan's from another's.
Grok 4.7 is too slow for routine work and not capable enough for hard work. That leaves no good coding use for the $30 plan. Its chat, image and video tools do not change our recommendation. Skip it.
Pros
Grok chat, images and video in one product
Cons
Too slow for routine work
Not capable enough for hard work
Best for Nobody.
How far it goes: we have not measured this plan
We grade Grok CLI on the $100 SuperGrok Plus plan - its scorecard is the closest evidence.
Allegretto gives K3 more credits and parallel agents. The shared limit still runs out before ordinary work is done. K3 has also regressed in our use since release. We do not recommend it.
Pros
Good visual design
Long context
Parallel agent tools
Cons
The limit drains too quickly
K3 has gone backwards in our use
Best for Nobody.
How far it goes: we have not measured this plan
The K3 subject sat out the latest weekly grading run, so there is no current scorecard; see the leaderboard for graded alternatives.
Our pick: Claude Max 5x. Claude Max 5x gives Opus 5.5 five times the Pro allowance. That is enough to keep several agents running through the day. Fable is included for the hardest planning, but I don't think using it is even the play on this plan. Move here when you repeatedly hit the Pro limit.
$100 gives Opus 5.5 five times the Pro allowance. Opus 5.5 is light enough on the limit to keep several agents running through the day. Fable is included for the hardest planning, but Opus 5.5 now covers most of it. Move here when the Pro limit stops you.
Pros
Five times the Pro allowance
Room for several Opus 5.5 agents
Sonnet 5.5 for smaller things and simple tasks
Fable for the hardest planning
Cons
Pro covers most single-agent days for $80 less
Best for People who keep hitting the Claude Pro limit. Start on Pro first.
How far it goes: at least 15 times a week, and more than 100 for the cheapest model
Our whole real-world issues and spec planning pass runs at least 15 times a week, and more than 100 for the cheapest model on this plan, depending on which model you run.
That works out at about under 23c to $1.84 of the subscription per run, at the list price.
Measured on this plan, at <1% to 8%/1w of its allowance per run. A share belongs to the plan it was measured on, so we never subtract one plan's from another's.
$99 gives K3 five times the agent credits of Allegretto. The usage limit still runs out too quickly. K3 has regressed in our use since release. More credits do not make it worth buying.
Pros
Five times the Allegretto agent credits
Good visual design
Cons
The usage limit drains too quickly
K3 has regressed since release
Best for Nobody.
How far it goes: we have not measured this plan
No graded Kimi subject at the moment; see the leaderboard for graded alternatives.
$100 gives Codex one weekly usage bucket with no 5-hour session window. You get every OpenAI model, including Astra. Buy it only if Plus's 5-hour windows are blocking you from doing more work.
Pros
No 5-hour session window
Every OpenAI model, including Astra
Luna for high-volume small tasks
Cons
No Opus 5.5 or Fable
Best for Codex users blocked by the Plus 5-hour windows.
How far it goes: we have not measured this plan
Same Codex CLI as ChatGPT Plus with five times its allowance - the Plus scorecards are the closest evidence.
$100 buys more Gemini usage and a larger Google bundle. It does not improve Antigravity or the available models. Start with the $20 AI Pro plan and upgrade only if its usage limit blocks your work.
Pros
More Gemini usage
A larger Google services bundle
Cons
The same rough Antigravity agent
The same models as the $20 plan
Best for Google AI Pro users blocked by its usage limit.
How far it goes: we have not measured this plan
Same Antigravity CLI as AI Pro with a bigger allowance - the AI Pro scorecards are the closest evidence.
$100 buys generous Grok limits, 1080p video and priority access. We rarely use Grok enough to need those limits. The coding model is still Grok 4.7. It is too slow for routine work and not capable enough for hard work. We do not recommend this plan.
Pros
Generous limits
1080p video
Priority access
Cons
Little reason to use enough Grok to need the limits
The same Grok 4.7 coding model
Best for Nobody.
How far it goes: about 7 times a week
Our whole real-world issues and spec planning pass runs about 7 times a week on this plan.
That works out at about $3.45 of the subscription per run, at the list price.
Measured on this plan, at 15%/1w of its allowance per run. A share belongs to the plan it was measured on, so we never subtract one plan's from another's.
Top $199 to $300/mo
Our pick: Claude Max 20x. Claude Max 20x is Max 5x with room for many agents running all day. Opus 5.5 does most of the work and Fable takes the hardest planning. The other $200 plans do not run Opus 5.5. The 20x applies to the 5-hour window. The weekly limit is only about twice the $100 plan.
$200 is the largest Claude allowance, with Opus 5.5 and Fable through Claude Code, the best CLI agent. That is room for many agents running all day. Opus 5.5 does most of the work. Fable is there for the hardest planning. Buy it only if the Max 5x limit is blocking your work.
Pros
The largest Claude allowance
Many Opus 5.5 agents all day
Sonnet 5.5 for smaller things and simple tasks
Fable for the hardest planning
Cons
Max 5x covers most days for $100 less
Best for People who keep hitting the Max 5x limit. Start on a cheaper plan first.
How far it goes: we have not measured this plan
Same models and harness as Claude Max 5x with more allowance - the Max 5x scorecards are the closest evidence.
$199 gives K3 ten times the agent credits of Allegretto. The usage limit still runs out too quickly. K3 has regressed in our use since release. More credits do not make it worth buying.
Pros
The largest Kimi credit pool
Good visual design
Cons
The usage limit drains too quickly
K3 has regressed since release
Best for Nobody.
How far it goes: we have not measured this plan
No graded Kimi subject at the moment; see the leaderboard for graded alternatives.
$200 gives Codex one weekly usage bucket with no 5-hour session window. It has four times the allowance of Pro 5x. You get every OpenAI model, including Astra. Buy it only if the Pro 5x weekly limit is blocking your work.
Pros
No 5-hour session window
Four times the Pro 5x allowance
Every OpenAI model, including Astra
Luna for high-volume small tasks
Cons
No Opus 5.5 or Fable
Best for Codex users blocked by the Pro 5x weekly limit.
How far it goes: we have not measured this plan
Same Codex CLI as ChatGPT Plus with twenty times its allowance - the Plus scorecards are the closest evidence.
$200 buys the highest Gemini usage and the largest Google bundle. It does not improve Antigravity or the available models. Buy this only if the lower Google plans cannot provide enough usage.
Pros
The highest Gemini usage
The largest Google services bundle
Cons
The same rough Antigravity agent
The same models as the $20 plan
Best for People who need maximum Gemini usage and will use the Google bundle.
How far it goes: we have not measured this plan
Same Antigravity CLI as AI Pro with a far bigger allowance - the AI Pro scorecards are the closest evidence.
The $100 plan already has more Grok usage than we can find a use for. We have no idea what anyone would do with the $300 plan. xAI does not publish a usage multiple. Coding still runs on Grok 4.7. We do not recommend this plan.
Pros
Includes X Premium+
Cons
No use case beyond the $100 plan
No published usage multiple
The same Grok 4.7 coding model
Best for Nobody.
How far it goes: we have not measured this plan
Same Grok CLI as SuperGrok Plus with a bigger allowance - the SuperGrok Plus scorecard is the closest evidence.
Verdicts last reviewed Oct 1, 2026, after the most recent weekly grading run. They, the band picks and the prices are the quiz's versioned editorial catalog (v48), reviewed in git - never computed from the live leaderboard. Grades and plan evidence come from published run data, and quota shares stay a share of each plan's own allowance; the honest cross-plan cost comparison is work per plan-dollar, on the leaderboard.