Which AI coding plans let buyers count what they get

AI coding plans now share price points, mostly $20, $100 and $200 a month, but not a unit of measure. Claude uses time windows, Cursor uses a weekly pool, Copilot uses credits worth a cent, Kiro uses its own credits, Bolt uses tokens, Lovable uses credits, and Augment uses dollars. Only the last group lets a buyer divide price by quantity. The others publish multiples of a base they never quantify, or ranges wide enough to be hard to plan around. The thesis of this analysis is that the metering unit matters more than the sticker price, because it decides whether a buyer can check a plan before paying. Even the countable plans leave a second variable, the cost per unit, that the buyer has to work out.

Every figure below was copied from vendor pricing and help pages. Claude, Cursor, GitHub Copilot, Kiro, Bolt, Lovable, Devin and Augment Code pages were checked on 2026-10-30, the Claude Ultra help article on 2026-10-31, and OpenAI pages on 2026-11-01. The Copilot billing page and the Cursor pricing pages were re-read on 2026-11-02 and show the same figures. Plans change often, so readers should confirm on the vendor's site before buying.

The same $100 buys different kinds of answer

The clearest way to see the split is to line up what each vendor says about a $100 monthly plan.

Nothing in the first three rows can be compared with the last five, or with each other. A buyer cannot say whether 5x Pro on Claude is more than 15x Plus on Cursor, because neither Anthropic nor Cursor says how big its base is. The $200 tier repeats the pattern: Claude Ultra at 10x Pro weekly, Cursor Ultra at 50x Plus, ChatGPT Pro at $200, Devin Max, Kiro Power at 10,000 credits and Lovable Pro at 800 credits.

Of the tools tracked, only Copilot, Kiro, Bolt, Lovable and Augment publish a countable monthly unit. Claude and Cursor publish only multiples, Devin publishes nothing countable, and OpenAI publishes per-model message ranges for Plus and Business.

Multiples of a hidden base

Claude's design is the most explicit about not giving numbers. Usage resets on a rolling five-hour session window, and paid plans add weekly limits that reset at a fixed time assigned to the account. Web, desktop, mobile and Claude Code draw from one pool. Anthropic publishes no fixed message count. It says usage depends on conversation length and complexity, attached files, the model, effort level and features used. Pro costs $20 and gets more usage than Free every five hours, but Anthropic no longer states a multiple of Free for Pro. The base of the whole ladder is therefore described only as "more than Free."

Above that base, the ladder is clean arithmetic. Max at $100 gives 5x Pro every five hours and every week, plus a separate weekly limit for Fable. Ultra at $200 drops the five-hour limit for verified accounts and gives 10x Pro weekly. Team standard seats at $25 give 1.25x the Pro per-session allowance, and premium seats at $125 give 6.25x. Against Pro's $20, those prices are 1.25 times and 6.25 times as large. The multiples therefore track price one for one, which suggests that Claude's tiers mostly sell more of the same thing in proportion. The open question is how much the thing is, and Anthropic does not say.

Cursor works the same way with a weekly pool. Every paid plan includes a pool for Cursor's own models (Grok 4.7, Grok 4.6, Grok 4.5 and Composer 2.5), and different models draw from it at different rates. Cursor publishes no request or credit counts. The tiers are Plus at $8 (the base), Premium at $30 (4x), Super at $100 (15x) and Ultra at $200 (50x). Here the multiples do not track price. Premium costs 3.75 times the Plus price for 4 times the pool. Ultra costs 25 times the Plus price for 50 times the pool. Per dollar, Ultra's pool is twice the size of Plus's, but a buyer still cannot say what the pool holds. A pricing ladder that rewards the biggest spender is easy to read, but the reward is measured against a base that is undisclosed.

Devin (Windsurf) is the limiting case. Each paid plan, Pro at $20 and Max at $200, gets an allowance that refreshes daily and weekly. Cognition publishes no credit, request or message counts, and extra usage is bought at API pricing. There is no unit to count, and none to compare.

Ranges and a coming switch at OpenAI

OpenAI is more specific than Anthropic or Cursor, and the specifics show how much a "number" can still leave open. Plus ($20) and Business ($25 per user per month, billed monthly, 2+ users) get Codex local-message estimates per five hours, given as a range per model:

Each range spans roughly eight to eleven times from bottom to top. OpenAI calls them "not fixed message limits." The count depends on the model, task size and complexity, and whether work runs locally or in the cloud, and cloud tasks may use more of the allowance than local messages. Weekly limits may also apply. A buyer planning a month of work from these figures is planning from a spread, not a quantity.

Other multipliers sit on top. Codex Fast mode uses included usage at 2.5x (2x on purchased credits), and Ultrafast at 8x (6x on credits). Image generation uses limits about 3 to 5 times faster. So the same plan delivers a different number of tasks depending on a mode setting. For Pro plans ($100, $200 and $500 a month), OpenAI says there is currently no five-hour limit, though weekly limits may still apply. That is a statement about a limit that is absent, not a quantity that is present.

The larger change is scheduled. OpenAI says that starting in December the usage allowance on Plus, Pro and Business will be shared across ChatGPT and Codex and measured in tokens, with Instant chat not drawing from it. It has not published per-plan token amounts. A move to tokens could make the allowance countable, but only if the amounts are published. This suggests the December figures are the thing to watch, and until they appear the token switch gives buyers nothing to count. Readers should confirm the amounts on OpenAI's site.

OpenAI's credit pricing already hints at what token metering looks like. After reaching limits, Plus and Pro users can buy credits billed per 1M tokens per model. For GPT-6.1 Sol the rate is 50 credits per 1M input tokens, 2.5 per 1M cached input tokens and 250 per 1M output tokens. Output costs five times as much as fresh input, and cached input one-twentieth. A call with 1M fresh input and 0.1M output would cost 50 plus 25, or 75 credits. The mix of work, not the number of messages, sets the bill.

Countable units still need a unit price

Five tools publish a monthly unit, and each one rewards a division that the vendor's page does not do for the buyer.

Kiro gives 50 credits per dollar on every paid tier: $20 for 1,000, $40 for 2,000, $100 for 5,000 and $200 for 10,000. That is $0.02 per plan credit at every tier, so a bigger plan buys no discount. Add-on credits cost $0.04 each, twice the plan rate. Plan credits do not roll over, while add-on credits roll over until they expire 12 months after purchase. Credits are metered to two decimal places with a 0.01 minimum. Simple prompts can cost less than 1 credit, complex ones such as spec tasks cost more, and models consume credits at different multipliers relative to Auto (1.0x). The count is exact, but the cost of a given task is not known in advance.

Copilot gives the most precise definition, since 1 AI credit equals $0.01. Pro at $10 has 1,000 base plus 500 flex credits (1,500 total). Pro+ at $39 has 3,900 base plus 3,100 flex (7,000 total). Max at $100 has 10,000 base plus 10,000 flex (20,000 total). Base credits match the subscription price, so Max's base is worth $100. At a cent per credit, the totals work out to $15, $70 and $200 of credit. That arithmetic is StackVersus's. The catch is that GitHub says the flex allotment may change over time. At Pro+, 3,100 of the 7,000 credits are flex, so the headline total is the less certain figure. Code completions and next edit suggestions do not use credits and are unlimited on paid plans. Included credits reset at 00:00:00 UTC on the first day of each calendar month. Business ($19 per user) adds 1,900 credits per user to a shared pool, and Enterprise ($39) adds 3,900. Paid plans get a 10% discount on model costs with auto model selection.

Lovable is linear on Pro: $25 for 100 credits, $100 for 400 and $200 for 800, all $0.25 per credit. Top-ups cost $0.30 per credit on Pro, and Business starts at $50 for 100 credits ($0.50 per credit). One balance covers building, chatting, hosting the built-in backend, AI features in deployed apps and connectors. A buyer can count credits, but what a credit buys depends on what the project does.

Bolt counts tokens, and the price per million falls with size in the Help Center examples. Pro starts at $25 for 10M tokens ($2.50 per million). The examples of 26M at $50 and 55M at $100 work out to about $1.92 and $1.82 per million. The unit is exact, but Bolt says most usage comes from syncing project files, so larger projects use more tokens per message. The same plan therefore yields fewer messages as a project grows. Unused paid tokens roll over and stay valid for two months.

Augment is the simplest on paper: Standard at $20 includes $20 of usage, and Business at $100 includes $100. But LLM usage is billed at the provider's API list price plus a flat 40% service fee. If the fee counts against the pool, $20 would cover about $14.29 of provider list-price usage. That is inference, and the vendor's billing page should settle it.

Overage is where the real price shows

The included allowance is the headline, but the overage rule shows what the vendor thinks a unit is worth. These rules vary more than the plan sizes do.

Claude lets paid users turn on usage credits billed at standard API rates when a limit is reached. Claude Enterprise includes no allowance at all: all tokens are billed at standard API rates, with a 20-seat minimum. Max and Ultra include $100 and $200 a month in API credits, but those work only on the Claude Platform, not in Claude Code or the Claude apps. Someone who buys Max for Claude Code gets no use from that credit there.

Cursor takes a different route to the same place. On Premium and above, third-party models such as Claude, GPT and Gemini sit outside the pool and are billed at API list price from the first request, out of prepaid credit. Free and Plus cannot use them. Teams and Enterprise pay an extra Cursor Token Rate of $0.25 per million tokens on eligible third-party requests. So the Cursor multiples apply only to Cursor's own models, and the larger tiers do not stretch the budget for the models many buyers name first.

Copilot charges $0.01 per credit beyond the pool, and paid extra usage is enabled by default, though admins can disable it. For a team that has not checked that setting, spend can run past the included credits. Kiro enterprise overage at $0.04 per credit is disabled by default, the opposite default. Free Kiro users cannot buy add-ons at all.

This suggests a practical reading. A plan with a countable unit and a safe default lets a buyer cap the worst case. A plan with a multiple of an unknown base and an open-ended overage at API rates makes the worst case hard to bound.

What to do

Pick the metering unit before the plan, and treat a plan with no countable monthly unit (Claude, Cursor's own pool, Devin, and OpenAI at least until it publishes token amounts) as a trial, not a budget line. For countable plans, divide price by units first: Kiro's $0.02 versus $0.04 add-on gap, Lovable's $0.25 versus $0.30 top-up, and Copilot's base versus flex split show where the headline overstates the guaranteed amount. Check the overage default before enabling a team: Copilot's paid extra usage is on unless an admin turns it off, and Kiro's enterprise overage is off. Finally, confirm every figure on the vendor's page on the day of purchase, because these pages change, OpenAI's token amounts are still unpublished, and GitHub says its flex credits may change.

Sources

• Claude pricing
• What is the Max plan? (Claude Help Center)
• What is the Ultra plan? (Claude Help Center)
• What is the Team plan? (Claude Help Center)
• Codex pricing (OpenAI)
• Understanding ChatGPT and Codex usage limits (OpenAI Help Center)
• Cursor models and pricing
• GitHub Copilot billing for individuals
• GitHub Copilot billing for organizations
• Kiro pricing
• Kiro add-on credits
• Bolt tokens (Help Center)
• Lovable plans and credits
• Augment Code pricing
• Devin pricing
• StackVersus AI plan limits tracker

Originally published on StackVersus, where prices and plan limits are checked against vendor pages and dated.